Systematic Literature Review on Test Case Selection and Prioritization: A Tertiary Study

Software testing is undertaken to ensure that the software meets the expected requirements. The intention is to find bugs, errors, or defects in the developed software so that they can be fixed before deployment. Testing of the software is needed even after it is deployed. Regression testing is an inevitable part of software development, and must be accomplished in the maintenance phase of software development to ensure software reliability. The existing literature presents a large amount of relevant knowledge about the types of techniques and approaches used in regression test case selection and prioritization (TCS&P), comparisons of techniques used in TCS&P, and the data used. Numerous secondary studies (surveys or reviews) have been conducted in the area of TCS&P. This study aimed to provide a comprehensive examination of the analysis of the enhancements in TCS&P using a thorough systematic literature review (SLR) of the existing secondary studies. This SLR provides: (1) a collection of all the valuable secondary studies (and their qualitative analysis); (2) a thorough analysis of the publications and the trends of the secondary studies; (3) a classification of the various approaches used in the secondary studies; (4) insight into the specializations and range of years covered in the secondary texts; (5) a comprehensive list of statistical tests and tools used in the area; (6) insight into the quality of the secondary studies based on the seven selected Research Paper Quality parameters; (7) the common problems and challenges encountered by researchers; (8) common gaps and limitations of the studies; and (9) the probable prospects for research in the field of TCS&P.


Introduction
The key to today's technological advance is programming and the continuous delivery of software. When software is released to users, it faces new challenges of meeting unforeseen user requirements, unexpected inputs, ever-growing market competition, and new and changing user demands. These issues all have to be handled after the software development phase is over and the software is in its maintenance phase. Then, the software either becomes obsolete, new software has to be developed (which is very costly in terms of computational time and resources needed), or the existing software has to be updated. Updating software has a clear budget and time constraints, and pressure to fulfill the desired modification goals. Hence, the updating process involves retesting the modified software to maintain its correctness and accuracy. This part of software testing is known as regression testing. Officially, "regression testing is performed between two different • Research question: The research question (RQ) of an SLR is specific and related to empirical studies' outcomes. The RQ of a SM is general and relevant to research trends. • Search process: Whereas the search process of an SLR is defined by the research question, the search process of SM is defined by the research topic. • Search strategy requirements: Whereas the search strategy requirements of an SLR are incredibly stringent, a mapping study's search strategy requirements are less stringent. All previous studies must be found for SLRs. • Quality evaluation: Quality evaluation is crucial for SLRs, but is not essential for mapping studies. • Results: Results of SLRs are answers to specific research questions. Results of SM are categories of papers related to a research topic.
When searching for research texts on online digital libraries, it is possible to miss out on essential texts that might not match the 'searched string'. One possible way to search such texts is snowballing. "Snowballing refers to using the reference list of a paper or the citations to the paper to identify additional papers. However, snowballing can benefit from looking at the reference lists and citations and complementing it with a systematic way of looking at where papers are referenced and cited. Using the references and the citations respectively are referred to as backward and forward snowballing" [16]. When performing surveys or reviews, this is a vital search procedure to avoid missing essential texts. Wohlin [16] explained the detailed procedure and guidelines for performing snowballing in surveying the literature for particular topics. Many researchers are taking advantage of this research strategy to conduct surveys. In addition to the SM techniques, snowballing was also used in searching the relevant studies in this paper.
TCS&P has attracted substantial research attention [17]. The increasing industrial demand for TCS&P techniques is a major driving force behind the research [18]. The advances in software technology and the rising complexity of systems have forced the software to be modified and retested. The amount of available time and cost for this retesting is usually minimal. Thus, the research for quicker, more accurate, and newer techniques to fulfill the changing technological demands is inevitable. It is essential to develop a new technique to justify its need: Doe such a method already exist? If it does exist, what are the possible shortcomings? Can the technique be updated to satisfy the requirements? Secondary studies are performed on existing literature to obtain a better answer to such questions. As the area of TCS&P has also generated numerous secondary studies, a tertiary study is therefore needed. This can provide a higher-level catalog of the research conducted in the area. This tertiary study is conducted by following the procedures used to perform a SM [8]. We used the guidelines provided by [8] because it provides detailed guidelines for carrying out a SM in the area of software engineering and is a highly cited paper (having more than 3000 citations). This study was also inspired from and supported by another tertiary study in the area of software testing [14]. We detected over 50 secondary studies in the field of TCS&P. Based on the inclusion and exclusion criteria discussed in later sections, only 22 secondary texts were selected for our survey; they were published from 2010 to 2020.
Contributions of this survey include: 1.
To present all the valuable secondary studies available online in the field of TCS&P in one place for quick referral.

2.
To analyze the basic publication information of these texts in the form of publishing journals, years, and online databases to be looked for. 3.
To classify the secondary studies based on the review approach used as SLRs, mappings, or survey/reviews. 4.
To analyze the breadth of search in the 22 studies. 5.
To provide a quick reference to the statistical tests and tools being used in TCS&P, with a reference to the study in which they have been used. 6.
To provide a detailed assessment of the quality of secondary research being conducted in TCS&P based on seven chosen Research Paper Quality parameters. 7.
To support the research community with a consolidation of advances in TCS&P, describing possible shortcomings in the existing texts and the probable future scope and likely publishing targets in the area.
The paper is structured as follows. Section 2 presents the adopted research methodology. Data extraction and summarization are presented in Section 3. The results and discussion and the conclusions drawn from this survey are provided in Sections 4 and 5, respectively.

Research Methodology
The research process was inspired by the guidelines [8] for performing SM in software engineering. The process begins with the research questions to be answered through an exhaustive literature review. The research questions then help in defining the Research String used to search the relevant texts in the different online databases. After relevant texts are identified, they are examined according to the exclusion criteria at multiple levels. Textbased exclusions take into account the inclusion and exclusion criterion pre-determined for the research. Based on these, the final texts are selected to be considered for review. The following subsections explain the details of the implementation of these steps.

Research Questions
The research questions were described and categorized into five parts. The categories then were further divided to provide the associated research questions. These questions form the basis of the current survey analysis of the studies. The details of the development of research questions are as follows: RQ 1. To get basic information of available secondary texts in TCS&P.

RQ 1.1: What is the distribution of secondary texts on TCS&P over the different online databases?
This RQ will enhance the knowledge of online databases in which secondary studies can be quickly found.

RQ 1.2:
What is the evolution of the number of secondary studies published in TCS&P over the years?
The expansion/compaction of the research field on TCS&P can be acknowledged by increasing/decreasing secondary studies in the area. New secondary studies are performed only when the existing ones have become obsolete or show missing aspects in their analysis. In both cases, researchers work in secondary research areas for which high demand is maintained in the research community. The analysis of publication numbers over time reveals the expansion of TCS&P secondary studies.

RQ 1.3:
Which are the key publishing journals for TCS&P secondary studies? Secondary texts are generally larger than the primary studies and involve a different research approach from those who propose new ideas. Therefore, not all conferences and journals support the publication of secondary studies. The answer to this RQ will provide possible publication targets for researchers working on secondary texts in the future. RQ 2. To study the characteristics of secondary studies.

RQ 2.1: Which research approaches have been used in secondary studies?
Several researchers have applied multiple techniques for secondary reviews on the existing texts. Different guidelines for different procedures have been presented [6,19]. We categorized the research approaches for performing secondary studies into three types: (1) systematic literature reviews, (2) mappings, and (3) reviews/surveys. Analyzing which texts use which approach and how these are distributed over time will help answer the RQ.

RQ 2.2:
What are the focus and the range of years covered in the secondary studies?
The analysis of the range of years of publication of the selected primary studies is an indicator of the length of coverage of secondary studies. It provides an insight into the variety of primary studies considered in the analysis. Surveys performed over similar ranges should have included similar texts that are related to the surveyed topic. It also helps future researchers search for primary studies that lie outside the already searched ranges.

RQ 2.3: How many texts form the basis of research for the selected secondary studies?
The extent of studies can also be judged depending on the number of chosen primary studies for conducting the survey. The number of primary studies chosen can vary widely for similar survey topics depending on the selected research questions. However, the number of mappings should be comparable for surveys on similar topics conducted for primary studies chosen from the same range of years. Comparisons and analysis of data in a secondary study may have used specific statistical tests.

RQ 5.2:
What is the extent of evolutionary techniques in secondary studies?
Evolutionary techniques have recently experienced a significant increase in use. These techniques are under continuous development and are used in all research fields concerning optimization problems. Thus, it is highly relevant to analyze coverage of evolutionary techniques in the TCS&P field. This will help researchers to discern if new or additional techniques can be applied in the area.

Defining the Search String
Given the research questions modeled in the previous section, our study's main topics of interest included surveys on selection and prioritization in the field of regression testing. Therefore, the search string included several spellings and synonyms valid for these topics. Search conditions were merged with the logical operators 'OR' and 'AND'. The SLR conducted by Kitchenham [7] on other existing SLRs inspired the formulated search string. Different online databases require the articulation of different search strings, depending on the advanced search options they provide. Table 1 (next section) summarizes the details of the search string and the databases. Streamlining the search results required advanced searching options for finding the terms in the abstract of the studies, given the large pool of available literature.
Moreover, we considered only English texts published in journals, conference proceedings, workshops, or book chapters. The range of search for our survey was from 1997 until May 2020. The beginning year was chosen as 1997 because the previous seminal studies [20,21] indicate that the research in the area of TCS&P began in this year. However, the first chosen secondary study in the field was published in 2008. (("Abstract": Prioritization OR "Abstract": selection OR "Abstract": prioritisation) AND ("Abstract": "test" OR "Abstract": "testing")) AND ("Abstract": "survey" OR "Abstract": "review" OR "Abstract": "mapping")

Conducting the Search
The search for secondary studies in TCS&P was accomplished in June 2020 following the PRISMA guidelines [28]. Table 1 summarizes the records published in online databases via the formulated search strings. It also shows the applied filters, the range of years of searched texts, and the number of studies resulting from the search.
In addition to these texts, additional literature in white papers, industry reports, and blogs are not present in these databases, because they are non-peer-reviewed texts. Such research is referred to as grey literature. Garousi et al. [29] propose a method for searching texts from grey literature. Due to the risks involved in studying grey literature, we restricted ourselves to technical reports and workshop papers. Kitchenham [7] used the same approach in her SLR on SLRs.
Additionally, we also considered the process of snowballing (see Section 1) by performing forward and backward snowballing of the citations and references, respectively, from selected studies found in online databases [16]. Thus, we identified studies that the search string may have missed. Twenty-four texts were initially identified due to snowballing and the grey literature search. Figure 1 shows the step-by-step procedure followed for conducting the search process based on the PRISMA ("Preferred Reporting Items for Systematic Reviews and Meta-Analyses") statement [28]. After the initial title-based search, we added abstract-based and text-based exclusions, resulting in 55 secondary studies in TCS&P. Then, we applied inclusion and exclusion criteria (Table 2), considering the research questions presented above. After all the steps, the final set included 22 texts for analysis. Table 2 summarizes the five inclusion and seven exclusion criteria that were applied to the first set of 55 contributions leading to the final set of 22 selected secondary studies. The most critical aspect was the quality check of the possible relevant studies. The exclusion criteria EC5, EC6, and EC7 were the basis for the rejection of studies. Many texts collected primary studies but did not include a proper analysis; these were excluded. For example, [30] is a systematic review on test case prioritization; however, we could find no analysis at all in the study. Other such studies were [31][32][33]. We also found a small number of secondary studies with an incomplete set of surveyed texts, and these were also excluded from the current study [34][35][36][37][38][39]. A survey on many-objective optimization was presented in [40]; the study was sound and effective but found to be outside the scope for the current study. Catal presented 10 best practices in test case prioritization, but the study did not present any analysis other than APFDs [41].  Table 2 summarizes the five inclusion and seven exclusion criteria that were applied to the first set of 55 contributions leading to the final set of 22 selected secondary studies. The most critical aspect was the quality check of the possible relevant studies. The exclusion criteria EC5, EC6, and EC7 were the basis for the rejection of studies. Many texts collected primary studies but did not include a proper analysis; these were excluded. For example, [30] is a systematic review on test case prioritization; however, we could find no analysis at all in the study. Other such studies were [31][32][33]. We also found a small number of secondary studies with an incomplete set of surveyed texts, and these were also excluded from the current study [34][35][36][37][38][39]. A survey on many-objective optimization was presented in [40]; the study was sound and effective but found to be outside the scope for the current study. Catal presented 10 best practices in test case prioritization, but the study did not present any analysis other than APFDs [41].

Inclusion and Exclusion Criteria
Although secondary texts that are not SLRs may not include all the relevant studies in the field, we rejected the studies that did not mention more than five relevant journal texts published in the surveyed range of years. The major reason for this is the inconsistency in the results with missing relevant information. In one of the cases, the authors presented the survey at a conference [42] and the detailed SLR was subsequently published in a journal [20]. In this case, considering EC7, specifically for the two surveys on TCS techniques by Engstrom et al. [20,42], the detailed text from the journal publication [20] was included in our selection.

IC1
Relevance to TCS&P. EC1 Texts that are not directly relevant to TCS&P, including secondary texts in software testing not specifically mention TCS&P.

IC2
Secondary studies such as surveys, reviews, SLRs and mappings are included. EC2 Primary texts proposing a new approach, or validating existing ones, or empirical comparison studies.

IC3
Texts published in journals, conferences, workshops, and technical reports are considered.

EC3
Texts found from magazine articles, theses, abstracts, posters, keynotes, doctoral symposiums, and non-peer-reviewed texts from blogs and comments.

IC4
Texts should be available in the English language. EC4 Texts are not available in the English language.

IC5
Quality of the text based on evaluation of primary studies, breadth of survey and new addition to research field.

EC5
Texts that did not provide any kind of analysis for the selected primary studies.

EC6
Texts missing many important primary texts published in the range of years reviewed.

EC7
Texts representing mere repetition of available secondary studies already published and not adding any new knowledge to the research. Although secondary texts that are not SLRs may not include all the relevant studies in the field, we rejected the studies that did not mention more than five relevant journal texts published in the surveyed range of years. The major reason for this is the inconsistency in the results with missing relevant information. In one of the cases, the authors presented the survey at a conference [42] and the detailed SLR was subsequently published in a journal [20]. In this case, considering EC7, specifically for the two surveys on TCS techniques by Engstrom et al. [20,42], the detailed text from the journal publication [20] was included in our selection.

Data Extraction and Summarization
All the 22 selected secondary studies were thoroughly examined to extract details according to RQs. Table 3 shows a summary of the selected texts. Additionally, we gathered information about the following evaluation aspects:

2.
Publication Details: The name of the publishing journal or conference where the secondary text was published.

3.
Year of Publication: The year in which the research was published.

1.
Publication Type: The publication type was categorized into a journal paper, conference paper, or book chapter.

2.
Focus Field: Focus of any secondary text considered for the current SM is either TCS or TCP; a few studies considered both.

3.
Survey Type: We categorized the survey types as SLRs, mappings, or surveys/reviews. SLRs and mappings were defined in previous sections. A third category encompassed any other strategy such as literature reviews, surveys, analysis, and trends.

4.
Range of years covered in the survey: Range of years of primary texts considered in each of the secondary studies.

5.
The number of studies: The number of primary texts analyzed in each secondary study, which is an indicator of the length of coverage. 6.
Analysis aspects: The type of analysis performed in each of study.

Results and Analysis
The detail answers to the research questions using the gathered data are presented in the following points.

RQ1-Basic Information of Available Texts in TCS&P
The three parts of the question were answered as follows:

RQ 1.1:
What is the distribution of secondary texts on TCS&P over various online databases? Figure 2 depicts a pie chart representing the spread of the chosen 22 studies over the different online databases. ACM and Elsevier account for 70% of the selected studies, followed by IEEE Explore and Springer digital library, respectively. Only one survey from Wiley online library was found relevant to our SLR. Thus, there is a stronger presence of TCS&P in ACM and Elsevier.

S18
A.Bajaj, O.P.Sangwan [51] IEEE Access S19 N.Gupta, A.Sharma, M.K.Pachariya [58] IEEE Access S20 P.Paygude, S.D.Joshi [59] ICCBI'18 S21 J.Lima, S.Virgilio [60] Information and Software Technology S22 M.Cabrera, A.Dominguez, I.Bulo [61] SAC'20 1. Publication Type: The publication type was categorized into a journal paper, conference paper, or book chapter. 2. Focus Field: Focus of any secondary text considered for the current SM is either TCS or TCP; a few studies considered both. 3. Survey Type: We categorized the survey types as SLRs, mappings, or surveys/reviews. SLRs and mappings were defined in previous sections. A third category encompassed any other strategy such as literature reviews, surveys, analysis, and trends. 4. Range of years covered in the survey: Range of years of primary texts considered in each of the secondary studies. 5. The number of studies: The number of primary texts analyzed in each secondary study, which is an indicator of the length of coverage. 6. Analysis aspects: The type of analysis performed in each of study.

Results and Analysis
The detail answers to the research questions using the gathered data are presented in the following points.

RQ1-Basic Information of Available Texts in TCS&P
The three parts of the question were answered as follows:

RQ 1.2: What is the evolution of the number of secondary studies published in TCS&P over the years?
Growing surveys in the area reveal the growth in the research field of TCS&P. The need for new secondary studies arises from the obsolescence of existing ones or from missing aspects in the analysis. The 22 selected studies were published between Jan 2010 to May 2020. Figure 3 shows a line graph with the number of secondary studies published during this period per year. There was rapid growth in the number of surveys conducted in TCS&P after 2015. The number for 2020 is clearly lower as our study only considers the year until May; it is likely that the number for the remainder of 2020 would be consistent with the early trend. Five valuable and rigorous surveys in TCS&P were published in 2019. This demonstrates the growing interest of researchers in the field.
ing aspects in the analysis. The 22 selected studies were published between Jan 2010 to May 2020. Figure 3 shows a line graph with the number of secondary studies published during this period per year. There was rapid growth in the number of surveys conducted in TCS&P after 2015. The number for 2020 is clearly lower as our study only considers the year until May; it is likely that the number for the remainder of 2020 would be consistent with the early trend. Five valuable and rigorous surveys in TCS&P were published in 2019. This demonstrates the growing interest of researchers in the field.

RQ1.3:
Which are the key publishing journals for TCS&P secondary studies? Figure 4 shows the type of publication of studies. Almost 70% of studies were journal papers, and the remainder were conference communications or book chapters. The length of text of a survey, SLR, or mapping study is generally larger than a typical general primary study. Not all journals in the software engineering field publish secondary texts due to domain and size constraints. Table 4 lists the names of the ten journals and the number of surveys on TCS&P. This list is not exhaustive for journals that might publish secondary texts; it simply shows where the selected secondary texts were published, as possible targets of publication for future contributions of researchers.

RQ 1.3:
Which are the key publishing journals for TCS&P secondary studies? Figure 4 shows the type of publication of studies. Almost 70% of studies were journal papers, and the remainder were conference communications or book chapters. The length of text of a survey, SLR, or mapping study is generally larger than a typical general primary study. Not all journals in the software engineering field publish secondary texts due to domain and size constraints. Table 4 lists the names of the ten journals and the number of surveys on TCS&P. This list is not exhaustive for journals that might publish secondary texts; it simply shows where the selected secondary texts were published, as possible targets of publication for future contributions of researchers.
during this period per year. There was rapid growth in the number of surveys conducte in TCS&P after 2015. The number for 2020 is clearly lower as our study only considers th year until May; it is likely that the number for the remainder of 2020 would be consiste with the early trend. Five valuable and rigorous surveys in TCS&P were published 2019. This demonstrates the growing interest of researchers in the field.

RQ1.3:
Which are the key publishing journals for TCS&P secondary studies? Figure 4 shows the type of publication of studies. Almost 70% of studies were journ papers, and the remainder were conference communications or book chapters. The leng of text of a survey, SLR, or mapping study is generally larger than a typical general pr mary study. Not all journals in the software engineering field publish secondary texts du to domain and size constraints. Table 4 lists the names of the ten journals and the numb of surveys on TCS&P. This list is not exhaustive for journals that might publish secondar texts; it simply shows where the selected secondary texts were published, as possible ta gets of publication for future contributions of researchers.  Reviewing primary texts can be accomplished via various techniques. The most commonly used methods are SLR, systematic mappings, and non-systematic surveys/reviews. The difference in performing SLR and mappings is discussed in Section 1. These are systematic approaches with available guidelines to be followed for searching and selecting the studies for review. Due to the systematic approach, these techniques take a large amount of time and effort to complete the review process. Many reviews may not claim to include all the studies in the relevant field or to provide conclusions after analysis. Such texts are very helpful in providing insights into various aspects and the collection of primary texts available in the reviewed area. Moreover, they consume comparatively less time and effort. Many researchers have adopted this approach for performing literature reviews, surveying particular aspects, and analyzing and depicting trends. All of these texts were classified to the survey/review category. Figure 5 provides the classification of the reviewed 22 studies into three types: eight SLRs, three mappings, and one paper claiming to be comprise both an SLR and a mapping. Ten studies were surveys/reviews but not SLRs or mapping studies. Reviewing primary texts can be accomplished via various techniques. The most com monly used methods are SLR, systematic mappings, and non-systematic surveys/review The difference in performing SLR and mappings is discussed in Section 1. These are sy tematic approaches with available guidelines to be followed for searching and selectin the studies for review. Due to the systematic approach, these techniques take a larg amount of time and effort to complete the review process. Many reviews may not clai to include all the studies in the relevant field or to provide conclusions after analysis. Suc texts are very helpful in providing insights into various aspects and the collection of pr mary texts available in the reviewed area. Moreover, they consume comparatively le time and effort. Many researchers have adopted this approach for performing literatu reviews, surveying particular aspects, and analyzing and depicting trends. All of the texts were classified to the survey/review category. Figure 5 provides the classification the reviewed 22 studies into three types: eight SLRs, three mappings, and one paper claim ing to be comprise both an SLR and a mapping. Ten studies were surveys/reviews but n SLRs or mapping studies. Figure 6 depicts publishing trends for these three types of surveys in TCS&P. Pu lishing trends for mappings are marked as 0 or 1 in a year. There was an apparent increa in the number of SLRs and reviews/surveys in TCS&P after 2015. The growing interest the research community in the field of TCS&P can be inferred from these publicatio trends.

RQ2.2:
What are the focus and the range of years covered in the secondary studies? Figure 7 uses the radar plot to represent the same years covered by the studies. It shows the beginning and the ending year of respective searches conducted in the studies (identified as S1 to S22). It was found that all surveys or review papers (fitting our inclusion criteria) were published after 2010. Hence the outer circle, which represents the end year of the search, encompasses a decade (2010-2020). One of the studies (S22) was undertaken after another existing study; hence the search years ranged only from 2017 to 2019.
Surveys performed over comparable year ranges must include almost the same texts. Similarly, future researchers can explore the primary studies lying outside the outer plot presented in Figure 7.
The 22 studies either focused on TCS, TCP, or a combination of both. Figure 8 shows how the majority of the secondary texts (82%) focused on TCP, whereas only 32% focused on TCS.
Research in the area of TCP began in 1997 [3], whereas that of TCS began in 1988 [62]. The surveys conducted after these studies used information gathered in these papers, so they needed to cover a reduced number of years. As is evident from Figure 7, most of the surveys were been conducted over the last two decades (2000-2020).  S8 S9 S10 S11 S12 S13 S14 S15 S16 S17 S18 S19 S20 S21 S22

Star t Year
End Year

RQ 2.2:
What are the focus and the range of years covered in the secondary studies? Figure 7 uses the radar plot to represent the same years covered by the studies. It shows the beginning and the ending year of respective searches conducted in the studies (identified as S1 to S22). It was found that all surveys or review papers (fitting our inclusion criteria) were published after 2010. Hence the outer circle, which represents the end year of the search, encompasses a decade (2010-2020). One of the studies (S22) was undertaken after another existing study; hence the search years ranged only from 2017 to 2019.

RQ2.2:
What are the focus and the range of years covered in the secondary studies? Figure 7 uses the radar plot to represent the same years covered by the studies. It shows the beginning and the ending year of respective searches conducted in the studies (identified as S1 to S22). It was found that all surveys or review papers (fitting our inclusion criteria) were published after 2010. Hence the outer circle, which represents the end year of the search, encompasses a decade (2010-2020). One of the studies (S22) was undertaken after another existing study; hence the search years ranged only from 2017 to 2019.
Surveys performed over comparable year ranges must include almost the same texts. Similarly, future researchers can explore the primary studies lying outside the outer plot presented in Figure 7.
The 22 studies either focused on TCS, TCP, or a combination of both. Figure 8 shows how the majority of the secondary texts (82%) focused on TCP, whereas only 32% focused on TCS.
Research in the area of TCP began in 1997 [3], whereas that of TCS began in 1988 [62]. The surveys conducted after these studies used information gathered in these papers, so they needed to cover a reduced number of years. As is evident from Figure 7, most of the surveys were been conducted over the last two decades (2000-2020).  S9 S10 S11 S12 S13 S14 S15 S16 S17 S18 S19 S20 S21 S22 Surveys performed over comparable year ranges must include almost the same texts. Similarly, future researchers can explore the primary studies lying outside the outer plot presented in Figure 7.

Star t Year End Year
The 22 studies either focused on TCS, TCP, or a combination of both. Figure 8 shows how the majority of the secondary texts (82%) focused on TCP, whereas only 32% focused on TCS.

RQ2.3: How many texts form the basis of research for the selected secondary studies?
The number of chosen primary studies for conducting the survey also measures the scope of studies. Depending directly on the selected research questions, there can be significant variations in the number of primary studies selected for SLRs. However, mapping studies must include literature that exists in the relevant topic being reviewed. Thus, the number of primary studies selected by surveys on similar topics chosen from the same range of years must be comparable. Figure 9 plots the number of primary studies reviewed by the 22 studies (S1-S22) except for S3, where the total number of reviewed texts was unavailable. The number varied from the lowest (seven) to the highest (320) in primary texts for the remainder of the studies. Mappings recorded almost similar studies from a similar range of years, as expected. S22 reported a maximum of 320 primary texts and only in 3 years (2017-2019), although the specific links to each of the studies were missing. All other papers provided links to the primary texts. Thus, researchers can find almost all primary texts until May 2020 in the TCS&P field directly from these 22 studies.  Research in the area of TCP began in 1997 [3], whereas that of TCS began in 1988 [62]. The surveys conducted after these studies used information gathered in these papers, so they needed to cover a reduced number of years. As is evident from Figure 7, most of the surveys were been conducted over the last two decades (2000-2020).

RQ 2.3: How many texts form the basis of research for the selected secondary studies?
The number of chosen primary studies for conducting the survey also measures the scope of studies. Depending directly on the selected research questions, there can be significant variations in the number of primary studies selected for SLRs. However, mapping studies must include literature that exists in the relevant topic being reviewed. Thus, the number of primary studies selected by surveys on similar topics chosen from the same range of years must be comparable. Figure 9 plots the number of primary studies reviewed by the 22 studies (S1-S22) except for S3, where the total number of reviewed texts was unavailable. The number varied from the lowest (seven) to the highest (320) in primary texts for the remainder of the studies. Mappings recorded almost similar studies from a similar range of years, as expected. S22 reported a maximum of 320 primary texts and only in 3 years (2017-2019), although the specific links to each of the studies were missing. All other papers provided links to the primary texts. Thus, researchers can find almost all primary texts until May 2020 in the TCS&P field directly from these 22 studies.

RQ2.3: How many texts form the basis of research for the selected secondary studies?
The number of chosen primary studies for conducting the survey also measures the scope of studies. Depending directly on the selected research questions, there can be significant variations in the number of primary studies selected for SLRs. However, mapping studies must include literature that exists in the relevant topic being reviewed. Thus, the number of primary studies selected by surveys on similar topics chosen from the same range of years must be comparable. Figure 9 plots the number of primary studies reviewed by the 22 studies (S1-S22) except for S3, where the total number of reviewed texts was unavailable. The number varied from the lowest (seven) to the highest (320) in primary texts for the remainder of the studies. Mappings recorded almost similar studies from a similar range of years, as expected. S22 reported a maximum of 320 primary texts and only in 3 years (2017-2019), although the specific links to each of the studies were missing. All other papers provided links to the primary texts. Thus, researchers can find almost all primary texts until May 2020 in the TCS&P field directly from these 22 studies.    Table 5 shows the details of the statistical tests in each study. The details include the area or purpose of the test, the reference of the studies where it was used, and the number of times it was used in the 22 surveyed papers.    Table 5 shows the details of the statistical tests in each study. The details include the area or purpose of the test, the reference of the studies where it was used, and the number of times it was used in the 22 surveyed papers.    10 S1 S2 S3 S4 S5 S6 S7 S8 S9 S10 S11 S12 S13 S14 S15 S16 S17 S18 S19 S20 S21 S22  Figure 11 shows a doughnut graph to provide better insight into the application of statistical tests. The performance analysis of the techniques proposed by the researchers used most of the tests (approximately 74%). The graph depicted in Figure 12 clearly shows that ANOVA, Mann-Whitney, Vargha and Delaney, and Bonferroni tests were the most used in TCS&P. Table 4 provides a quick reference to the studies where the usage of these tests was recorded.

No. of Statistical Tests used in various Surveys
Appl. Sci. 2021, 11, x FOR PEER REVIEW 15 of 33 Figure 11 shows a doughnut graph to provide better insight into the application of statistical tests. The performance analysis of the techniques proposed by the researchers used most of the tests (approximately 74%). The graph depicted in Figure 12 clearly shows that ANOVA, Mann-Whitney, Vargha and Delaney, and Bonferroni tests were the most used in TCS&P. Table 4 provides a quick reference to the studies where the usage of these tests was recorded. Unfortunately, this analysis indicates a major gap in the usage of statistical tests in regression TCS&P. This provides an opportunity for researchers who could make future use of statistical tests for qualitative assessment, validation, and comparison of the techniques used in TCS&P.     Figure 11 shows a doughnut graph to provide better insight into the application of statistical tests. The performance analysis of the techniques proposed by the researchers used most of the tests (approximately 74%). The graph depicted in Figure 12 clearly shows that ANOVA, Mann-Whitney, Vargha and Delaney, and Bonferroni tests were the most used in TCS&P. Table 4 provides a quick reference to the studies where the usage of these tests was recorded. Unfortunately, this analysis indicates a major gap in the usage of statistical tests in regression TCS&P. This provides an opportunity for researchers who could make future use of statistical tests for qualitative assessment, validation, and comparison of the techniques used in TCS&P.   Unfortunately, this analysis indicates a major gap in the usage of statistical tests in regression TCS&P. This provides an opportunity for researchers who could make future use of statistical tests for qualitative assessment, validation, and comparison of the techniques used in TCS&P.

RQ 3.2:
Which tools have been used by the secondary studies? Table 6 summarizes the information about the open-source behavior, available download links, and the referring studies of the mentioned research tools.  References for the tools mentioned in Table 6 are given in Table 7.  A few studies mention other tools in addition to those mentioned in Table 5. The secondary study S2 refers to a text [88] that contains a small section of the survey of early tools developed for regression testing, listing their advantages and disadvantages. S6 mentions tools created to automate their own proposed techniques. Some of the tools, such as Vulcan, BMAT, Echelon, déjà vu, GCOV, Test runner, Winrunner, Rational test suite, Bugzilla, and Canatata++, were only mentioned in one study each. S8 mentions two tools: RTSEM and MISRA-C. S9 mentions three more tools: SPLAR, Feature IDE, and MBT.
As is evident from Figure 13, only 14% of the studies mentioned the tools used in the area of regression TCS &P. These tools are listed in Table 5, with the study in which they were referenced, the categorization/purpose of the tool (if mentioned), and the download link (if the tool is open source).
The downloadable links and referring study links can help future researchers to obtain a quick reference to the tools that have already benefitted the TCS&P field. Figure 14 graphically shows that 25% of the tools are used for analysis purposes, 21% of tools are used for providing code coverage information, 19% of tools are used for stipulating the mutation adequacy score, 8% of tools are used for delivering source code metrics, and the remainder are used for miscellaneous purposes. In general, the TCS&P area lacks the usage of standard tools, and the tools used are concentrated in the analysis of techniques and code coverage. Thus, this observation suggests there is scope for finding ways of standardizing the set of possible tools to encourage more automation in the area of regression TCS&P. suite, Bugzilla, and Canatata++, were only mentioned in one study each. S8 mentions two tools: RTSEM and MISRA-C. S9 mentions three more tools: SPLAR, Feature IDE, and MBT.
As is evident from Figure 13, only 14% of the studies mentioned the tools used in the area of regression TCS &P. These tools are listed in Table 5, with the study in which they were referenced, the categorization/purpose of the tool (if mentioned), and the download link (if the tool is open source). The downloadable links and referring study links can help future researchers to obtain a quick reference to the tools that have already benefitted the TCS&P field. Figure 14 graphically shows that 25% of the tools are used for analysis purposes, 21% of tools are used for providing code coverage information, 19% of tools are used for stipulating the mutation adequacy score, 8% of tools are used for delivering source code metrics, and the remainder are used for miscellaneous purposes. In general, the TCS&P area lacks the usage of standard tools, and the tools used are concentrated in the analysis of techniques and code coverage. Thus, this observation suggests there is scope for finding

RQ4. How Can the Quality of Secondary Studies Be Compared?
The categorization presented for various TCS&P techniques in the secondary studies was recorded and is plotted in Figure 15. The highest percentage of studies categorize the techniques based on the approach, whereas a few studies specifically chose genetic algorithms (GA), and 27% of the studies chose other unique categorization techniques.

RQ4. How Can the Quality of Secondary Studies Be Compared?
The categorization presented for various TCS&P techniques in the secondary studies was recorded and is plotted in Figure 15. The highest percentage of studies categorize the techniques based on the approach, whereas a few studies specifically chose genetic algorithms (GA), and 27% of the studies chose other unique categorization techniques.

RQ4. How Can the Quality of Secondary Studies Be Compared?
The categorization presented for various TCS&P techniques in the secondary studies was recorded and is plotted in Figure 15. The highest percentage of studies categorize the techniques based on the approach, whereas a few studies specifically chose genetic algorithms (GA), and 27% of the studies chose other unique categorization techniques. The categorization trend can be justified by the evident change in the upcoming techniques over time. In addition to this, a few studies focused on a specific area of TCS&P rather than on the general analysis of TCS&P. Therefore, categorization is not sufficient for quality assessment.

RQ4.1: How have the research questions been used in the secondary studies?
The different approaches used in secondary studies clearly highlight the importance of research questions when performing a secondary study. Hence, the research questions The categorization trend can be justified by the evident change in the upcoming techniques over time. In addition to this, a few studies focused on a specific area of TCS&P rather than on the general analysis of TCS&P. Therefore, categorization is not sufficient for quality assessment.

RQ 4.1: How have the research questions been used in the secondary studies?
The different approaches used in secondary studies clearly highlight the importance of research questions when performing a secondary study. Hence, the research questions (RQs) covered by secondary studies in TCS&P were analyzed. Figure 16 represents the number of RQs considered by the studies. Most of the studies mentioned between three and five RQs. The count of RQs includes the sub-questions of the RQs, as mentioned in the study. Publication trends determine the most found RQs. The majority of the secondary studies included RQs that are factual in nature. Only five studies analyzed the collected facts and discussed their findings from the analysis. No peculiar trend could be seen in terms of increasing RQs or in the improvement in the quality of RQ.
Appl. Sci. 2021, 11, x FOR PEER REVIEW 20 of 33 (RQs) covered by secondary studies in TCS&P were analyzed. Figure 16 represents the number of RQs considered by the studies. Most of the studies mentioned between three and five RQs. The count of RQs includes the sub-questions of the RQs, as mentioned in the study. Publication trends determine the most found RQs. The majority of the secondary studies included RQs that are factual in nature. Only five studies analyzed the collected facts and discussed their findings from the analysis. No peculiar trend could be seen in terms of increasing RQs or in the improvement in the quality of RQ.  Figure 17 shows that 63 % of the surveys in TCS&P focused on illuminating the 'facts' of working in the area. Only 23% of the secondary studies also focused on the analysis part of the work accomplished in the area. This highlights the deficit in analytically surveying or reviewing the TCS&P field over time. This also motivated us to propose detailed analytical RQs as much as possible within the logical technical constraints and limitations of time and cost. 10 S1 S2 S3 S4 S5 S6 S7 S8 S9 S10 S11 S12 S13 S14 S15 S16 S17 S18 S19 S20 S21 S22 No. of RQ's Mentioned

Secondary Studies
No. of RQ's    Figure 17 shows that 63 % of the surveys in TCS&P focused on illuminating the 'facts' of working in the area. Only 23% of the secondary studies also focused on the analysis part of the work accomplished in the area. This highlights the deficit in analytically surveying or reviewing the TCS&P field over time. This also motivated us to propose detailed analytical RQs as much as possible within the logical technical constraints and limitations of time and cost.  Figure 18 indicates that the majority (77%) of the surveys clearly and substantially answered the defined RQs. The remainder of the surveys either lack a pre-defined objective for the study or did not substantially answer the defined RQs. Although 23% of the 0 1 S1 S2 S3 S4 S5 S6 S7 S8 S9 S10 S11 S12 S13 S14 S15 S16 S17 S18 S19 S20 S21 S22

RQ4.2:
What is the quality of the secondary studies performed in the area?
We worked with a total of seven Research Paper Quality (RPQ) parameters in order to provide a qualitative analysis of the secondary studies published in the area of regression TCS&P. These RPQs were designed to derive their inspiration from seminal studies [20,21] in the area of TCS&P. Table 8 lists the RPQ values allocated to the 22 surveyed studies during research and analysis.  Figure 18. Distribution of quality of research questions answered over the 22 studies for the survey.

RQ 4.2:
What is the quality of the secondary studies performed in the area?
We worked with a total of seven Research Paper Quality (RPQ) parameters in order to provide a qualitative analysis of the secondary studies published in the area of regression TCS&P. These RPQs were designed to derive their inspiration from seminal studies [20,21] in the area of TCS&P. Table 8 lists the RPQ values allocated to the 22 surveyed studies during research and analysis. RPQ1 (RQ) denotes whether the study mentions and answers the RQ or not. A study was assigned the value '1' if it was found to mention the RQs; else, it was assigned the value '0'. RPQ2 (RQ Quality) represents the quality of the RQ research in the secondary study. Value '0' means that no research questions appeared in the secondary study, so they were not answered. Secondary studies having RQs framed to provide only descriptive analysis of the existing data, without providing exploratory, inferential, predictive, or diagnostic analysis of the primary studies, were assigned the value '0.5'. Value '1' means high-quality RQs were specified and answered. RPQ3 (Future Prospects): A study that mentions the scope of future prospects or gives directions to contribute further to the area received the value of '1' whereas value '0' was assigned to the remainder. RPQ4 (Statistical Test): If statistical tests used in the area by different researchers were present in the secondary study, the value was '1'; otherwise, the value was '0'. RPQ5 (Tools available): If the tools used in the area of TCS&P were mentioned in the study, then the value '1' was assigned; else the value '0' was assigned. RPQ6 (Detailed Analysis): Adopts value '1' for a study that provided a detailed and indepth analysis of the research work accomplished in the area of TCS&P. A study with a value '0.5' included only limited descriptive analysis of the research work such as publication years and details. Value '0' was assigned to a study that was a mere collection of facts and did not contain any analysis of the research conducted in the area. RPQ7 (Novel Contribution): Value '1' means that the study made a novel contribution in the analysis of the research conducted in TCS&P. Value '0.5' appears when the contribution or analysis was a mixture of some novel findings and the repetition of some findings already available in the literature, and value '0' implies that the contribution did not include any significant or novel analysis of the research. Total (RPQ) sums the values of the RPQ to provide a comprehensive qualitative analysis of the secondary studies. Our qualitative analysis shows that S6, S18, and S21 contributed most significantly to the area, and S2, S4, S5, S7, S11, S12, and S13 also contributed significantly to the area.
Even after a significant increase in the number of secondary studies in TCS&P, it was not accompanied by an improvement in quality as measured by the seven RPQ parameters. This suggests an opportunity for improving the quality of secondary studies in the future. Here we present categories of TCS&P techniques in a tabular manner (Table 9) to highlight the kinds of techniques available in the area. The second column indicates the studies which mention the technique. The techniques (Coverage-based, History-based, Requirement-based, Model-based, and Fault-based) were categorized into over 10 surveys; however, all of these are old techniques. In addition, we observed that Evolutionary techniques (which include Search-based, GA-based, ACO-based, ABC and BA, and hill climbing) saw a significant rise in development after 2015. Hence, to keep up with the current research, it would be informative to identify the trend in research on evolutionary techniques. The subsequent RQ analyzes and addresses this aspect. Table 9. Category of TCS&P techniques mentioned in the studies.

RQ 5.2:
What is the extent of Evolutionary techniques in secondary studies?
Recent research has witnessed the advent of many evolutionary techniques, such as GA, ACO, and hill climbing. New evolutionary techniques are continuously being developed and applied to various optimization problems, including software engineering problems [89][90][91]. TCS&P is also an optimization problem and hence has benefitted from various evolutionary techniques. Table 10 summarizes the details of the evolutionary techniques used in TCS&P in the 22 secondary studies. Twelve evolutionary techniques were detected as part of the secondary studies. Table 8 shows the details of the studies where they appear, with quick references to the primary studies in which the techniques were applied to TCS&P. Table 10. Details of evolutionary techniques surveyed in the 22 secondary studies. (GAgenetic algorithms, PSO-particle swarm optimization, AVM-alternating variable method, EAevolutionary algorithms, HC-hill climbing, ACO-ant colony optimization, SD-string distance, BA-bacteriological algorithm, BCO/ABC-bee colony optimization/artificial bee colony).

5
AVM(1+1) S9(2016) [78] 6 Evolutionary Algorithm S9(2016) [78] 7 Hill Climbing S10(2016), S14(2018) [81] 8 ACO S12(2017), S14(2018), S15(2018), S17(2019), S20(2019) [21,[138][139][140][141][142][143][144] 9 String Distance S12(2017) [145] 10 Bacteriological Algorithm S15(2018) [146] 11 Bee Colony Optimization S15(2018) [141] 12 ABC S19(2019) [147]  Notably, the number of secondary studies mentioning evolutionary techniques also rose after 2015 (see Figure 20). This is evident from the fact that evolutionary techniques increased in number for TCS&P after 2015 ( Figure 20) and that the number of secondary studies conducted in TCS&P also rose post-2015 ( Figure 3). It can also be easily inferred that the extensive use of evolutionary techniques in TCS&P is increasing. Notably, the number of secondary studies mentioning evolutionary techniques also rose after 2015 (see Figure 20). This is evident from the fact that evolutionary techniques increased in number for TCS&P after 2015 ( Figure 20) and that the number of secondary studies conducted in TCS&P also rose post-2015 ( Figure 3). It can also be easily inferred that the extensive use of evolutionary techniques in TCS&P is increasing. Another relevant finding is that GA is the most commonly applied evolutionary technique in TCS&P (see Figure 21), and has been used in 16 secondary studies. GA and its variants are popular in obtaining solutions to many other real-world problems [148,149]. The two next highest ranking are ACO and PSO. Additional new techniques are being developed while earlier approaches continue to be used. All these findings provide incentive for researchers to work on evolutionary techniques in TCS&P.  Another relevant finding is that GA is the most commonly applied evolutionary technique in TCS&P (see Figure 21), and has been used in 16 secondary studies. GA and its variants are popular in obtaining solutions to many other real-world problems [148,149]. The two next highest ranking are ACO and PSO. Additional new techniques are being developed while earlier approaches continue to be used. All these findings provide incentive for researchers to work on evolutionary techniques in TCS&P.

Threats to Validity
Both secondary and tertiary studies are susceptible to threats to validity [150]. We followed the categorization provided by Ampatzoglou et al. [150] for classifying the possible threats to validity present in the case of SLRs or SMs. We also mention the steps taken to mitigate the effects of these threats in the following section.

1.
Study Selection Validity: One of the main threats in the case of SLRs or SMs is the elicitation of relevant studies. We formulated a search string for automated selection from six databases. The string was manually tested with a trial-and-error technique to check if it found all the well-known studies. In addition, we also performed snowballing to gather any study missed while performing automated retrieval of studies. However, it is possible that we missed a small number of studies that used different terminology to describe their secondary study (such as 'analysis' or 'study synthesis'or 'study aggregation', or any similar phrases or strings). Moreover, we excluded grey literature in the area by assuming that good quality grey literature is generally available in conference/journal papers. It is also desirable to use Scopus and WoS databases for the retrieval of relevant studies. In our case, we could not obtain access to these databases, so we used Google Scholar and other seminal databases (IEEE, Wiley, Science Direct, Springer, and ACM) to gather the relevant studies.

2.
Data Validity: Data validity threats occur when performing data extraction or analysis of the extracted data. Individual bias may be a factor when locating relevant facts and subjective data from the gathered studies. The data extraction and analysis in our case was initially performed by one author. Then, another author verified the process of extracting the data for analysis to provide a check on the extracted data. The remaining authors then reviewed the overall analysis.

3.
Research Validity: This threat is associated with the overall research design. In order to mitigate this threat, we tried to follow a research methodology that is very well formulated and recognized by researchers in the area. We thereby followed the guidelines provided by Petersen et al. [8], which specify a thorough approach to carrying out an SM specifically in the area of software engineering. Its high citation count (more than 3000 citations) ensures that these guidelines are recognized by renowned researchers working in the field. Our study was also inspired by and supported by another tertiary study in the area of software testing [14].

Conclusions
Our SM adopted quantitative and qualitative analyses to assemble the research findings in the area of TCS&P. The secondary studies (reviews or surveys conducted on primary studies) found in the field of TCS&P were reviewed systematically to obtain an overall picture of the recent findings in the area. This work primarily presented a general data analysis of the papers, and a detailed data analysis of the included studies corresponding to the formulated RQs. The studies were thoroughly analyzed to gather information, and to comprehensively tabulate results and findings. This study can provide a quick guide to researchers working in the area of TCS&P for clear insight into the trends of work undertaken in the area, in addition to the tools, statistical tests, limitations, and probable prospects.
One goal of our study was the detailed analysis of the 22 selected secondary studies to find common findings and limitations. Although we identified very few common findings, we tried to compile results and generate conclusions from them. These are stated as follows: • Five of the studies [S1, S2, S13, S15, S16] report that programs downloaded from SIR (Software-artifact Infrastructure Repository) are the most commonly used benchmark programs for the evaluation of the techniques in the area.
Four studies [S13, S14, S16, S21] noted that APFD (Average Percentage of Faults Detected) is the most preferred metric in the assessment of techniques in regression TCS&P. Three studies [S7, S18, S19] recorded that GA (genetic algorithm) is the most commonly used approach employed by the researchers working in the area. Four studies [S8, S11, S14, S19] concluded that coverage-based selection and prioritization is the most preferred criterion, whereas S15 reports that there is a paradigm shift from coverage-based to nature-inspired or search-based approaches. S21 reports that a history-based system is the most preferred criterion by researchers working in regression TCS&P.
We also identified some common problems, research gaps, or limitations identified by the analysis of the surveys.

•
The studies S1, S10, and S11 observed that regression TCS&P techniques had been applied to limited and usually small test benches (as reported by [S4]).

•
The studies S1, S3, and S19 showed that there is a lack of application of regression TCS&P techniques to testing for non-functional requirements. • Two studies, S7 and S16, reported a lack of common tools used in TCS&P. The studies S10 and S16 state that the prioritization techniques available in the literature do not provide the execution order of the prioritized test cases. • S19 and S21 highlight the fact that there is a gap in the literature on how to achieve tighter integration between the regression techniques and debugging techniques. • S7 highlights that most of the regression TCS&P techniques proposed have focused on a particular application domain or are context-specific; hence, assessment of the superiority of a technique over the others is not possible. It also notes the gap in the lack of techniques for mobile applications or web services, and the lack of regression TCS&P techniques for software using GUI or complex domain software. • S10 stated that most of the existing prioritization techniques are evaluated using APFD. This measurement, however, suffers from a large number of constraints in practice.
The analysis of the studies enables a collective report of the following prospects of future work in the area, which may also overcome limitations and gaps.

•
Our analysis indicates a major gap in the usage of statistical tests in regression TCS&P. This opens an opportunity to researchers who could make use of statistical tests for qualitative assessment, validation, and comparison of the techniques used in TCS&P in the future. • S7, S1, and S3 recommended the use of TCS&P techniques for software domains such as SOA, web services, and model-based testing, and embedded, real-time, and safety-critical software.
• S4, S5, and S18 suggest priority to public datasets rather than proprietary ones. • S1, S7, S11, and S19 support the usage of multicriteria-based TCS&P techniques. In addition, S1 and S3 recommend the usage of regression TCS&P techniques for model-based testing. • S10 states that it is necessary to optimize the execution of test cases, mainly due to the cost of individual test cases rather than the interest in total cost. • S14 recommends prioritizing the order of multiple test suites rather than test cases and suggests measuring the efficiency of TCP techniques based on actual time spent on fault detection. • S16 suggests the usage of APFDC to explicitly consider test execution time.

•
Finally, S21 provides a good presentation of prospects in regression TCS&P.
The significant contributions of the accomplished study are: (1) presents highlights of the publication trends and a list of the popular journals and conferences relevant to the area of TCS&P; (2) enumerates the statistical tests used in the area; (3) comprehensively provides a list of tools used in the area, with the source; (4) lists the test benches and metrics commonly used and the most frequent approaches; and (5) lists the limitations of the research conducted and the prospects for future work in the area.
The results of this SM are subject to the following limitations. Firstly, the secondary studies selected for the study are limited according to the adopted inclusion criteria and the specified RQs. Secondly, the search criteria of the selected papers were limited to only those in the English language. These conditions were essential for the feasibility of our SM. In summary, this SM reports the recent progress in the area of TCS&P to provide insight into previous research and to identify prospects for further work in the area.

Conflicts of Interest:
The authors declare no conflict of interest.