Location-Based Analyses for Electronic Monitoring of Parolees

: This study analyses the spatio-temporal pattern of parolees using electronic monitoring, where the developed spatial framework supports the Environmental Criminology concepts such as crime patterns or crime attractive locations. A grid-based solution for spatio-temporal analyses is introduced to ensure the anonymity of the parolees. In order to test these developed concepts, the Istanbul Metropolitan Area was selected as the pilot study area. Following the developed concepts of the Crime Pattern Theory, a spatial framework was designed. A novel grid-based weighted algorithm for the most attractive areas was generated via spatial point-of-interest data and a conducted survey among police o ﬃ cers. The designed framework and the spatio-temporal analyses were carried out for 77 parolees using geostatistical methods. The major ﬁndings of the study are (a) 24-hour trajectories of each parolee was monitored; (b) the most attractive grids within the city were deﬁned; and (c) for each parolee, the entrance time to the grid and the time spent within that grid were reported and analyzed. This study is a preliminary study for spatio-temporal detection of parolees’ trajectories, where location-based analyses serve well. This study aims to aid decision-makers to better monitor the parolees and justify the beneﬁts of surveillance.


Introduction
The concept of electronic monitoring of parolees has emerged by integrating technology to probation applications. By definition, parole "refers to criminal offenders who are conditionally released from prison to serve the remaining portion of their sentence in the community" [1], where the probation is a period when a criminal must behave well and not commit any more crimes to avoid being sent back to prison. It is the method used for monitoring the convicted prisoners or detainees due to prosecution or suspicion of offence. The monitoring process could easily benefit from information technologies, namely location-based services and spatial data analyses, where such applications could highly benefit from mature geospatial surveys. The use of electronic monitoring systems enables the spatial and temporal monitoring of the perpetrators of crime; besides, it intersects with the place (or situation) and the offender dimensions of the four basic dimensions of the crime [2]. Hence, environmental criminology, where the problem has a spatial nature, examines the crime phenomenon in terms of the "where" and "when" questions. Researchers and decision-makers in environmental criminology perform spatial and temporal analyses to determine the criminal patterns and make conceptions regarding the crime itself [3]; however, the full benefits of location-based analyses have yet to be discovered.
With 15 million inhabitants, Istanbul is the most populous city in Turkey and Europe. Considering that Turkey's population is 80 million, Istanbul hosts almost one-fifth of the country's population even though it is not the capital of the country. It has 39 districts, and the Istanbul Strait connects Asia to Europe. This historic city, currently considered as the cultural and economic center of the country, was founded a long time ago and it spreads over 5343 km 2 [16].
According to the police and gendarmerie statistics, 383,750 incidents occurred in 2016 and 445,408 incidents occurred in 2017 in Istanbul Province [17]. The most important reasons for the selection of Istanbul as the study area are because it has the highest population density and the highest number of electronic monitoring possibilities by GNSS.
Electronic monitoring, which began in 1964 at Harvard University [18], had not been used by judicial authorities to track criminals until 1983. In 1983, Jack Love, the judge of Albuquerque in New Mexico, U.S., sentenced three criminals to house control under an electronic monitoring order [19,20]. Today, electronic monitoring systems are widely used in many countries [21,22]. The studies carried out by Işık [23] in Turkey, Hucklesby and his colleagues [24,25] in Europe, Renzema [26] in the USA and Di Tella and his colleague [27] in Argentina on usage rates by countries were some of the studies carried out on this subject since 1983.
In electronic monitoring systems, the active GNSS systems, which are among the operating principles of electronic monitoring, are designed to continuously receive and send signals of the location and time information of the monitored person. Today, GNSS (such as GPS, Galileo, BeiDou, and GLONASS) systems are extensively used due to technological developments [28]. Electronic monitoring systems communicate with monitoring centers through software. GNSS Electronic monitoring systems use global navigation satellite systems with geographical information systems that store and manage spatial data [29] by using advanced data mining techniques [13]. GNSS electronic monitoring system data can be used for the detection of behavior patterns [30,31]. In this study, data obtained from GNSS-supported active electronic monitoring systems were applied. Thus, highly valuable spatial data of people who have previously committed a crime and are now electrically monitored were used in the analysis.
In Turkey, electronic monitoring is carried out by the Department of Probation under the General Directorate of Prisons and Detention in the Ministry of Justice. This system in Turkey is applied for the fulfilment of some judicial supervision measures imposed in place of detention before the verdict, the fulfilment of some optional punishment imposed in place of imprisonment after the verdict, and for the supervision and monitoring of released criminals within the society and following their conditional release from the prison. According to data released in 2017 [32], electronic monitoring has begun in 2012 in Turkey, and has been applied to 27,313 cases until April 19, 2017. Of the total electronic monitoring cases, only 1785 cases were monitored using GNSS tracking systems. There are no home detention and victimized unit cases that were monitored by using the same system. In addition, there was no data and document on how many parolees commit crimes after being released in Turkey.
In this study, the selected cases covered 77 people; 5 of them were female and 72 were male. The total number of records of electronic monitoring data was 1,919,587 within Istanbul. The average number of records for all parolees was 26,288. The data was collected between the years 2014 and 2017. According to the records, parolees were labelled as follows: 37 persons (48%) for sex crimes, 14 persons for assault (18%), 14 persons for theft (18%), and 12 persons for murder (16%). Hence, no specific groups were therefore targeted. This is the full dataset within the Ministry of Justice GNSS tracking system. The system was initialized in 2014, where all records were retrieved until 2017. The original data obtained from the Ministry of Justice included 124 parolees. To obtain better analysis results, 2 parolees tracked in less than 7 days and 45 of the parolees that belonged to any of the crime group that has less than 10 people have been eliminated.
Point-of-interest (POI) data was retrieved from the OpenStreetMap (OSM) [33] website. This information was used to determine the locations for crime generators and attractors in environmental criminal theories. Accordingly, POIs representing bars, night clubs, pubs, ATM's, banks, stadiums, sports centers, gyms, playgrounds, airports, coach stations, terminals, hotels, motels, hostels, guesthouses, universities, schools, colleges, hospitals, cafes, beverages, grocery stores, car parks, courthouses, shopping malls, dormitories, public buildings, bus stations, railway stations, ferries, terminals, picnic sites, campsites, toilets, department stores, supermarkets, restaurants, fast-food outlets, graveyards, kindergartens, sanctuaries, museums, cinemas, and theatres were selected. The number of POIs was 32,901. In the study, the coordinates and usage of the POIs were used as the main data to calculate the most attractive areas, which were generally evaluated as sources for the crime. In the study, the breakdown of POIs according to location types was as follows: 3490 restaurants, 2829 bus stops, 2577 cafes, 2375 pitches, and 2199 supermarkets.
The attractive locations for criminals may be assessed in environmental criminology theories with the local characteristics in different societies. In this study, a survey was conducted through Google Forms [34] to measure the probability of a crime occurring in places that may be attractive to the criminals. This survey was conducted with 51 policemen who were selected among those that were considered to be more in contact and familiar with crime and the criminal. In the questionnaire, the 5-point Likert scale was used to measure the likelihood of all types of crime occurring in POIs. Scoring of the locations in the range of 1 (strongly disagree) to 5 (strongly agree) was requested from the respondents.

Methodology Used
The framework for the analysis and data processing was determined and shown in Figure 1. Along with the framework, some steps were identified as the explanatory data analysis (EDA) of the data [35][36][37]. The most important part was the spatio-temporal analysis of the data to be analyzed through geographic information systems and R programming because of its spatial nature. Due to the sensitivity of the data, information on parolees should be kept anonymous and the study area needs to be generalized. Thus, 1 km × 1 km grids were generated, and the survey subsequently carried out. Tools used for the required analysis were as follows: the open-source PostgreSQL geographic database, ESRI ArcGIS Software [38], and R: A Language and Environment for Statistical Computing [39], for data mining with the spatio-temporal analysis methods. terminals, picnic sites, campsites, toilets, department stores, supermarkets, restaurants, fast-food outlets, graveyards, kindergartens, sanctuaries, museums, cinemas, and theatres were selected. The number of POIs was 32,901. In the study, the coordinates and usage of the POIs were used as the main data to calculate the most attractive areas, which were generally evaluated as sources for the crime. In the study, the breakdown of POIs according to location types was as follows: 3490 restaurants, 2829 bus stops, 2577 cafes, 2375 pitches, and 2199 supermarkets. The attractive locations for criminals may be assessed in environmental criminology theories with the local characteristics in different societies. In this study, a survey was conducted through Google Forms [34] to measure the probability of a crime occurring in places that may be attractive to the criminals. This survey was conducted with 51 policemen who were selected among those that were considered to be more in contact and familiar with crime and the criminal. In the questionnaire, the 5-point Likert scale was used to measure the likelihood of all types of crime occurring in POIs. Scoring of the locations in the range of 1 (strongly disagree) to 5 (strongly agree) was requested from the respondents.

Methodology Used
The framework for the analysis and data processing was determined and shown in Figure 1. Along with the framework, some steps were identified as the explanatory data analysis (EDA) of the data [35][36][37]. The most important part was the spatio-temporal analysis of the data to be analyzed through geographic information systems and R programming because of its spatial nature. Due to the sensitivity of the data, information on parolees should be kept anonymous and the study area needs to be generalized. Thus, 1 km x 1 km grids were generated, and the survey subsequently carried out. Tools used for the required analysis were as follows: the open-source PostgreSQL geographic database, ESRI ArcGIS Software [38], and R: A Language and Environment for Statistical Computing [39], for data mining with the spatio-temporal analysis methods.
The framework was divided into three main categories, namely, data input, data analysis, and data output. They are presented in Figure 1, and explained below.

Data Input
The data input category consists of the collection of data. In this study, the electronic monitoring data of the parolees who had previously committed crimes in Istanbul, have probation and electronic  The framework was divided into three main categories, namely, data input, data analysis, and data output. They are presented in Figure 1, and explained below.

Data Input
The data input category consists of the collection of data. In this study, the electronic monitoring data of the parolees who had previously committed crimes in Istanbul, have probation and electronic monitoring decisions with GNSS, and also the POI data of Istanbul were used. Data on electronic monitoring were obtained by special permission taken from the Department of Probation of the General Directorate of Prisons and Detention Centers of the Ministry of Justice of Turkey. The Ministry of Justice of Turkey has provided electronic monitoring data anonymously within the scope of the personal data security. The data included spatial information, such as "ID, latitude, longitude, and time", as well as demographic information, such as "probation name, article type, crime type, gender, and age". In the data received, the number of parolees who were electronically monitored with GNSS was 77. The first data date from the received data was 16 February, 2014, and the last data date was 20 February, 2017. The temporal sampling of the parolees was one minute. The total number of records received was 1,919,587, this after eliminating the data belonging to those areas outside the provincial borders of Istanbul. The duration of the monitoring of people who were being electronically monitored varied. This process changed according to the decision made by the Department of Probation.
While the abovementioned processes were progressing, a survey study was designed and implemented apart from these procedures.

Data Modelling
After the data input category, the first step in the data modelling category was the transfer of the electronic monitoring data and POI data to the geographic information systems software with the geographic database created in PostgreSQL. The POI data and then the electronic monitoring data were transferred from the geographical database to the GIS environment. In OSM, most of the POI were stored as an areal (polygon) object. Hence, such data were transformed into points and merged into the GIS environment.

Data Analysis
At this stage, after converting all the different spatial data into nodal data, grids were created to provide better representation of these data that has been analyzed. These grids both make the analysis more understandable and help to settle ethical concerns about the privacy of individuals. For this reason, 1 km × 1 km grids were created to cover the city of Istanbul in the GIS environment. Furthermore, these 1 km 2 grids have been evaluated as an adequate spatial resolution to conduct similar spatial algorithms. After the grids were created, the numbers of POIs falling into each grid were counted. These grids were formed independently. Based on the survey results of these grids, the most intense ones were selected.
In the survey study, it was aimed to learn the crime rate coefficients of the participants by taking into account the POI information obtained from OSM. Therefore, because of the multiplication of coefficients gathered from the survey study with the number of POIs in each grid, the total score was calculated. This score has been used to determine the attractiveness level of the grids. For instance, when all of the police officers gave 5 points for nightclubs, each nightclub was multiplied by 5 points. In case of two nightclubs in one grid, 10 points from the nightclubs were added to the total score. An example of the total score calculation is shown in Figure 2. As can be seen in Figure 2, there existed 1 park, 5 pharmacies, 1 railway station, 2 sanctuaries, and 2 schools that can be defined as places that attracted crime within the 1 * 1 km 2 grid drawn with a red frame. The coefficients were calculated with the Equation (1), and they were assumed to be values given in Equation (2).
Total Score = (1 park × 2.71) + (5 pharmacy × 3.08) + (1 railway station × 2.75) + (2 sanctuary × 2.2) + (2 school × 3.18) = 31.62 The map of Istanbul, which was obtained as a result of the calculations in the relevant example shown in Figure 2, was subjected to the necessary symbology and visualization procedures. In these symbology processes, visualization was based on the impression method [40] that depended on the standard deviation, used in density surface maps related to crime analysis.
All data sets obtained with GIS and the geographic database were transferred to the R programming environment. The "arcgisbinding" package [41] was used in the transfer process. Data discovery was performed with packages such as "tidyverse" [42] and "ggplot" [43] to be further used in the R programming environment. Another package used for spatio-temporal analysis in the R programming environment was "recurse" [44]. With the "recurse" package, it was possible to calculate the number of repeated entry times in certain areas and the time spent in the specified areas. The "sp" package [45,46] was used for spatial operations in the R programming environment. In this Total Score: Score to be used to determine the most attractive areas: a, b, c, . . . .z: Crime attractor and crime generator places I 1 , I 2 , I 3 , . . . . i n : As a result of the survey, the related crime type of the crime incidence; As can be seen in Figure 2, there existed 1 park, 5 pharmacies, 1 railway station, 2 sanctuaries, and 2 schools that can be defined as places that attracted crime within the 1 * 1 km 2 grid drawn with a red frame. The coefficients were calculated with the Equation (1), and they were assumed to be values given in Equation The map of Istanbul, which was obtained as a result of the calculations in the relevant example shown in Figure 2, was subjected to the necessary symbology and visualization procedures. In these symbology processes, visualization was based on the impression method [40] that depended on the standard deviation, used in density surface maps related to crime analysis.
All data sets obtained with GIS and the geographic database were transferred to the R programming environment. The "arcgisbinding" package [41] was used in the transfer process. Data discovery was performed with packages such as "tidyverse" [42] and "ggplot" [43] to be further used in the R programming environment. Another package used for spatio-temporal analysis in the R programming environment was "recurse" [44]. With the "recurse" package, it was possible to calculate the number of repeated entry times in certain areas and the time spent in the specified areas. The "sp" package [45,46] was used for spatial operations in the R programming environment. In this way, the calculation of the staying times in the grids with the repeated entry times to the most attractive areas was performed.

Data Output
Following the completion of the data analysis process in the GIS and R programming environments, the process of obtaining the results of these analyses started. In the output phase, tables and other outputs were obtained with the help of maps in the GIS and R programming environments.

Results and Discussion
Before conducting the spatial analysis, electronic monitoring data retrieved from the authorities were explored in detail.
According to the type of crimes, the average, minimum, and maximum ages were descriptively analyzed, and the obtained results achieved are shown in Figure 3. way, the calculation of the staying times in the grids with the repeated entry times to the most attractive areas was performed.

Data Output
Following the completion of the data analysis process in the GIS and R programming environments, the process of obtaining the results of these analyses started. In the output phase, tables and other outputs were obtained with the help of maps in the GIS and R programming environments.

Results and Discussion
Before conducting the spatial analysis, electronic monitoring data retrieved from the authorities were explored in detail.
According to the type of crimes, the average, minimum, and maximum ages were descriptively analyzed, and the obtained results achieved are shown in Figure 3. According to the dataset of the parolees, the average monitoring day per person was 66 days, where the highest and lowest number of monitoring days were 9 and 122, respectively. The records of the parolees according to the type of crimes and their total numbers are provided in Table 1. Most of the data were collected from people who committed sex crimes; secondly, assault; thirdly, murder; and fourthly, theft. Various spatial analyses have been conducted with this information, where some of them are provided in Figure 4. The initial analysis was to detect the activity pattern of paroles within the city. The activity patterns of parolees belonging to the four types of crimes were mapped between 10 a.m. According to the dataset of the parolees, the average monitoring day per person was 66 days, where the highest and lowest number of monitoring days were 9 and 122, respectively. The records of the parolees according to the type of crimes and their total numbers are provided in Table 1. Most of the data were collected from people who committed sex crimes; secondly, assault; thirdly, murder; and fourthly, theft. Various spatial analyses have been conducted with this information, where some of them are provided in Figure 4. The initial analysis was to detect the activity pattern of paroles within the city. The activity patterns of parolees belonging to the four types of crimes were mapped between 10 a.m. and 8 p.m. The basis of this time selection was the assumption that this time interval presented the busiest hours of routine everyday activities. There were also possibilities of crime being committed outside the selected hours. Nevertheless, we focused on the range of hours where mobility was most intense, and where the majority of the circulation occurred.  Hotspot analysis was also carried out as another spatial analysis. A total of 141 cells were determined as a hotspot with the confidence of 90% and above. The result of the hotspot analysis (Getis-Ord Gi*) of the parolees between 10 a.m. and 8 p.m. is shown in Figure 5. In this part of the study, only hotspots were considered. Thus, the cold spots obtained from the analysis have been excluded. The 1 km x 1 km grids were applied to generalize the detailed location information. Hotspot analysis was also carried out as another spatial analysis. A total of 141 cells were determined as a hotspot with the confidence of 90% and above. The result of the hotspot analysis (Getis-Ord Gi*) of the parolees between 10 a.m. and 8 p.m. is shown in Figure 5. In this part of the study, only hotspots were considered. Thus, the cold spots obtained from the analysis have been excluded. The 1 km x 1 km grids were applied to generalize the detailed location information.
Hotspot analysis was also carried out as another spatial analysis. A total of 141 cells were determined as a hotspot with the confidence of 90% and above. The result of the hotspot analysis (Getis-Ord Gi*) of the parolees between 10 a.m. and 8 p.m. is shown in Figure 5. In this part of the study, only hotspots were considered. Thus, the cold spots obtained from the analysis have been excluded. The 1 km x 1 km grids were applied to generalize the detailed location information. As a result of the survey, the crime attractiveness per location was assessed. Participants were asked to rate POIs concerning crime attractiveness, where a scale between 1 and 5 was used. Within the scale, 5 was the most attractive, whereas 1 was the least attractive. The results of the survey are illustrated in Table 2. According to the survey, the riskiest places were bars, nightclubs, and pubs. In line with the previous studies, this result showed that these places could play a significant role in the formation of the crime [47][48][49][50]. The second category were universities, while the third and fourth were cafes, as well as ATM's and banks, respectively, which are usually crowded. In contrast to the studies in which restaurants and public buildings were frequently considered crime attracting and crime generating places [51][52][53][54], these places were almost half more reliable than bars and nightclubs according to respondents.  As a result of the survey, the crime attractiveness per location was assessed. Participants were asked to rate POIs concerning crime attractiveness, where a scale between 1 and 5 was used. Within the scale, 5 was the most attractive, whereas 1 was the least attractive. The results of the survey are illustrated in Table 2. According to the survey, the riskiest places were bars, nightclubs, and pubs. In line with the previous studies, this result showed that these places could play a significant role in the formation of the crime [47][48][49][50]. The second category were universities, while the third and fourth were cafes, as well as ATM's and banks, respectively, which are usually crowded. In contrast to the studies in which restaurants and public buildings were frequently considered crime attracting and crime generating places [51][52][53][54], these places were almost half more reliable than bars and nightclubs according to respondents. The visualization of crime attractiveness grids by total scores is shown in Figure 6.  According to the survey conducted in this study, certain areas and environmental factors were considered as influential in the formation of criminals and the crime, but they were preventable and designable. The study showed that there was a greater intensification in some of the grids that were 1 km x 1 km in size, which were sufficient to reveal the patterns of the criminals in the intensified cells. There were 23 different cells higher than 400 points according to the weight of the total score.
After this stage, these 23 cells were used in all analyses in the R programming environment. A total of 54 of the 77 parolees have entered 2044 different times into these grids. The results of the analysis of how long the parolees spent time in the 23 cells, which were the most attractive areas, are shown in Figure 7. Those who committed sex crimes accounted for 52% of the total time spent inside the most attractive areas. It was observed that sex crimes parolees tended to go to the most attractive areas further than other types of crime parolees in the used dataset. Of those who committed murder, less than 30% spent their time inside the most attractive areas. According to the survey conducted in this study, certain areas and environmental factors were considered as influential in the formation of criminals and the crime, but they were preventable and designable. The study showed that there was a greater intensification in some of the grids that were 1 km × 1 km in size, which were sufficient to reveal the patterns of the criminals in the intensified cells. There were 23 different cells higher than 400 points according to the weight of the total score.
After this stage, these 23 cells were used in all analyses in the R programming environment. A total of 54 of the 77 parolees have entered 2044 different times into these grids. The results of the analysis of how long the parolees spent time in the 23 cells, which were the most attractive areas, are shown in Figure 7. Those who committed sex crimes accounted for 52% of the total time spent inside the most attractive areas. It was observed that sex crimes parolees tended to go to the most attractive areas further than other types of crime parolees in the used dataset. Of those who committed murder, less than 30% spent their time inside the most attractive areas. The time spent in each cell according to the type of crime is shown in Figure 8. In 19 of the 23 cells, parolees spent less than the mean of the time spent, and in 4 cells they spent more time. The parolees who committed sex crimes were seen especially in grids 7, 10, 11, and 18. As illustrated, people who committed murder were concentrated in the 7th grid. The result of the analysis of the time spent by the parolees who had electronic monitoring and the time passed after they have entered are shown in Figure 9. In this analysis, the lower limit was determined as 1 min and the upper limit was determined as 1440 min. In this way, short stays were excluded. The red line in the figure shows the mean of the time spent by entrance time. It was also noticed that those who committed sex crimes entered the attractive areas at 11:00 a.m. and have spent more time there. However, those who entered at 10.00 a.m. have spent comparatively less time.
One of the major findings, namely, varying opportunities depending upon the crime type, is also The time spent in each cell according to the type of crime is shown in Figure 8. In 19 of the 23 cells, parolees spent less than the mean of the time spent, and in 4 cells they spent more time. The parolees who committed sex crimes were seen especially in grids 7, 10, 11, and 18. As illustrated, people who committed murder were concentrated in the 7th grid. The time spent in each cell according to the type of crime is shown in Figure 8. In 19 of the 23 cells, parolees spent less than the mean of the time spent, and in 4 cells they spent more time. The parolees who committed sex crimes were seen especially in grids 7, 10, 11, and 18. As illustrated, people who committed murder were concentrated in the 7th grid. The result of the analysis of the time spent by the parolees who had electronic monitoring and the time passed after they have entered are shown in Figure 9. In this analysis, the lower limit was determined as 1 min and the upper limit was determined as 1440 min. In this way, short stays were excluded. The red line in the figure shows the mean of the time spent by entrance time. It was also noticed that those who committed sex crimes entered the attractive areas at 11:00 a.m. and have spent more time there. However, those who entered at 10.00 a.m. have spent comparatively less time.
One of the major findings, namely, varying opportunities depending upon the crime type, is also The result of the analysis of the time spent by the parolees who had electronic monitoring and the time passed after they have entered are shown in Figure 9. In this analysis, the lower limit was determined as 1 min and the upper limit was determined as 1440 min. In this way, short stays were excluded. The red line in the figure shows the mean of the time spent by entrance time. It was also noticed that those who committed sex crimes entered the attractive areas at 11:00 a.m. and have spent more time there. However, those who entered at 10.00 a.m. have spent comparatively less time.
focused on identifying concentrations of crime events and the prediction of crime locations. For example, Chainey et al. [55] have examined hotspot methods, including grids, to predict the spatial patterns of crime, where the crime location data was utilized. In this study, the hotspot method (Getis-Ord Gi*) was used to determine the highest values in the implemented grids, where the information was retrieved from the GNNS-based electronic monitoring of parolees. Furthermore, this spatial analysis incorporates POI and GNSS data, where, due to the sensitivity of data, a spatial gridding approach is utilized. Via the followed approach, the duration of staying in a specific type of location could be measured.
This study also used a wide variety of geo-statistical methods to analyze this interesting and unique data set for Turkey. However, several others could not be applied at this stage due to the limitations of the current dataset.

Conclusions
In this study, the added value of spatio-temporal analyses to environmental criminology was explored. It is possible to analyze and visualize the data of parolees and their places with the help of spatio-temporal analyses techniques and algorithms. Various categories of crime were investigated, where the major finding was that parolees tend to go to certain areas that are defined as the most attractive area. The spatial analyses assisted to understand crime and criminal behavior. It was also noticed that if those who committed crimes enter the attractive areas at certain times, they spent more time there. It is possible to evaluate that most attractive areas are places that are probably also important for parolees. It is observed that the most attractive areas in Istanbul are concentrated in the city center. This confirms that the places in which crime attractors and crime generators are located in is the most densely populated areas of the city. Similar valuable information can be presented to decision-makers by the collection of more spatial and temporal information on the subject through having a higher number of parolees with electronic monitoring, thus improving the spatio-temporal analyses.  One of the major findings, namely, varying opportunities depending upon the crime type, is also fully in line with another study on electronic monitoring (Rossmo et al. [13]). Accordingly, some types of crime need longer periods of planning than others. Another major finding, detection of hotspots via clustering analyses, has also contributed to revealing the spatial patterns of crime. Hence, a similar small number of studies regarding the spatial patterns of crime is supported. Such studies generally focused on identifying concentrations of crime events and the prediction of crime locations. For example, Chainey et al. [55] have examined hotspot methods, including grids, to predict the spatial patterns of crime, where the crime location data was utilized. In this study, the hotspot method (Getis-Ord Gi*) was used to determine the highest values in the implemented grids, where the information was retrieved from the GNNS-based electronic monitoring of parolees. Furthermore, this spatial analysis incorporates POI and GNSS data, where, due to the sensitivity of data, a spatial gridding approach is utilized. Via the followed approach, the duration of staying in a specific type of location could be measured.
This study also used a wide variety of geo-statistical methods to analyze this interesting and unique data set for Turkey. However, several others could not be applied at this stage due to the limitations of the current dataset.

Conclusions
In this study, the added value of spatio-temporal analyses to environmental criminology was explored. It is possible to analyze and visualize the data of parolees and their places with the help of spatio-temporal analyses techniques and algorithms. Various categories of crime were investigated, where the major finding was that parolees tend to go to certain areas that are defined as the most attractive area. The spatial analyses assisted to understand crime and criminal behavior. It was also noticed that if those who committed crimes enter the attractive areas at certain times, they spent more time there. It is possible to evaluate that most attractive areas are places that are probably also important for parolees. It is observed that the most attractive areas in Istanbul are concentrated in the city center. This confirms that the places in which crime attractors and crime generators are located in is the most densely populated areas of the city. Similar valuable information can be presented to decision-makers by the collection of more spatial and temporal information on the subject through having a higher number of parolees with electronic monitoring, thus improving the spatio-temporal analyses.