Measuring Urban Quality of Life Through Spatial Analytics and Machine Learning: A Data-Driven Framework for Sustainable Urban Planning and Development

Abstract

Today, the expansion of cities and rapid urbanization have led to unsustainable development and reduced quality of life (QOL) in urban ecosystems. This research aimed to establish a new framework for measuring QOL in a city by using spatial data and integrating the fuzzy analytical hierarchy process (FAHP) with support vector machine (SVM) methods. Four main components, including socioeconomic level, urban land use, urban environment, and natural environment; eleven subcomponents; and seventeen spatial indicators were defined. To produce quality-of-life maps of Mashhad City, the components, subcomponents, and indicators were integrated using weights obtained via the FAHP method. Then, SVM was applied to semi-automatically produce QOL maps. The results showed that Regions 2, 3, 4, 6, 10, and 11 displayed lower QOL scores, especially regarding the environmental and socioeconomic indicators. Regions 1 and 7, as well as Districts 0902, 0903, 0501, and 0502, showed average QOL in regard to the natural environment and socioeconomic indicators. Regions 8 and 12, along with District 0901 and Samen City, obtained better QOL scores in regard to nearly all indicators, except for access to land uses and the NDVI index. The results show that using the SVM method, a QOL map—with a kappa coefficient of 0.97 and an overall accuracy of 98%—can be successfully created with significant time, cost, and effort savings.

1. Introduction

The concept of quality of life (QOL) was initially articulated by the World Health Organization (WHO) in 1948, defined as a condition of comprehensive physical, mental, and social well-being rather than as simply the absence of illness and disability. The definition of QOL is multi-dimensional and may change over time, considering our changing society, cultural values, and individual perceptions of QOL relative to the environment [1]. Investigating QOL is essential because it is effective in predicting human behavior and provides happiness and satisfaction in their living environment. Therefore, to better understand QOL in a city or a settlement, its conditions should be measured by several appropriate indicators [2]. The authors of [3] suggested that physical and environmental conditions affect QOL. The authors of [4] define urban QOL as the level of an individual’s overall satisfaction with life. Moreover, the author of [5] describes it as the state of the environment in which people live. Most researchers agree that QOL is a multi-dimensional concept, including psychological, economic, social, and physical well-being [6].

QOL experienced by individuals residing in urban communities is shaped by the dynamics occurring within their natural and anthropogenic environments. Today, urban development and urbanization have led to the inappropriate expansion of cities. Therefore, QOL varies in different areas of a city [7]. In general, there are two main approaches to assessing QOL in cities. The first is the subjective approach, which is a perceptual and subjective criterion comprising evaluations of life and the living environment, as provided by residents or citizens. This approach includes data that are assessed by field surveys and questionnaire methods. In this method, all issues related to the access status, pollution, facilities, and services in the study area are posed through questionnaires, and the QOL is estimated based on the percentages of users’ answers. The strength of this approach is linked to the researcher’s experience in meeting the needs of the residents of the study area. Moreover, its weakness lies in the lack of accuracy and correctness of the data obtained from the questionnaires and surveys [8]. The latter type is an objective approach, which is based on different spatial data and decision-making methods, including the following:

• Statistical, official, governmental, and census data;
• Spatial data, such as geospatial information system (GIS) layers;
• Remote sensing data [9,10,11].

This approach allows for data and information comparison and does not depend on people’s understanding [8].

Statistical and census datasets have been used to check QOL. Usually, information is extracted from the census and statistical data to measure economic and social status. These data include employment and unemployment status, literacy, population size, poverty level, welfare facilities, and other social statuses. Statistical data must be location-oriented, so the information needs to be converted into spatial data [6,12,13,14]. GIS layers, such as information regarding building polygons, road maps, population, schools, etc., are appropriate for producing a QOL model. GIS data produced by national organizations are usually reliable for research studies. However, data revision occurs every few years, which is not ideal for annual studies. Therefore, researchers used other sources, such as data-driven remote sensing with the integration of GIS data [15,16,17,18,19]. As mentioned previously, remote sensing indicators are a good source for QOL studies. Advantages such as revisit time, different temporal and spatial resolutions, access to vast or distant areas, etc., make it popular for use in environmental studies. Spatial indicators such as vegetation, temperature, air pollution, and land cover products can be effectively used in QOL studies [6,7,14,20,21]. Based on the advantages of remote sensing and GIS data, it seems that integrating the environmental information from remote sensing and socioeconomic GIS-based data is a good solution for monitoring QOL. Spatial land use/land cover data can provide a more detailed urban spatial pattern, which plays an essential role in understanding issues related to QOL, such as urban infrastructure distribution and urban green space fairness [22].

Appropriate approaches must be employed to integrate remote sensing and GIS data to obtain QOL. Based on the literature, researchers used multicriteria decision-making (MCDM) methods and machine learning approaches to model QOL. Some researchers used MCDM methods, such as the analytic network process (ANP), FAHP, and AHP, to integrate factors to create a QOL map. The abovementioned methods are based on the use of weights for each indicator and their integration via mathematical models. In MCDM methods, experts play an essential role in assigning weights [16,18].

In contrast to these methods, machine learning approaches benefit from the use of computer models to solve a problem. In some studies, machine learning approaches such as support vector machine (SVM), random forests (RFs), and artificial neural networks (ANNs) were employed to make QOL maps [23,24,25]. Machine learning (ML) approaches are a subset of artificial intelligence. In machine learning, it is necessary to use some training samples to train mathematical models [26]. Based on previous studies, the use of ML or MCDM methods in preparing QOL maps has sparked interest. Combining these two methods can integrate artificial intelligence and expert knowledge, which is considered a gap in previous studies. This study aimed to establish a new framework for measuring QOL in Mashhad City using spatial data and integrating FAHP with SVM methods. So, an integration method based on MCDM and ML was introduced to produce a QOL map. Training data were obtained from the QOL map that was produced from the FAHP method and used to learn the support vector machine method. Four main components, including socioeconomic level, urban land use, urban environment, and natural environment, as well as eleven subcomponents and seventeen spatial indicators, were defined to create the desired framework.

2. Materials and Methods

2.1. Study Area

The study area is located in Mashhad, a city that has experienced rapid development due to high immigration rates, making it a suitable choice for this research. Figure 1 illustrates Mashhad’s location in Iran, with geographic coordinates 36.2972° N, 59.6067° E, along with its administrative divisions—13 regions and 34 districts.

• Region 1, one of the city’s oldest areas, is located in the city center. This region has commercial, administrative, medical, and academic centers.
• Region 2 has four districts. Districts 1 and 4 are adjacent to the city center, and Districts 2 and 3 are considered to be on the city’s outskirts. Districts 1 and 4 are old and busy, with commercial centers. Districts 2 and 3 are characterized by lower socio-cultural status.
• Region 3 comprises three districts: District 1, situated adjacent to the city center, experiences high foot traffic due to its proximity to the railway station. In contrast, Districts 2 and 3, located on the urban periphery, exhibit unregulated development and diminished socio-cultural standards.
• Region 4 includes three districts, with Districts 1 and 2 positioned near the city center, while District 3 lies on the outskirts.
• Region 5 similarly consists of three districts: District 1 is centrally located, whereas Districts 2 and 3 occupy peripheral zones.
• Region 6 also features three districts. District 1 is centrally situated, with the remaining two districts extending to the outskirts.
• Region 7 contains three districts: District 1 hosts the city’s airport, District 3 encompasses peripheral areas and an industrial park, and District 2 remains undescribed.
• In Region 8, District 1 lies near the historic city limits and houses commercial hubs and a major government hospital. District 3 includes a critical passenger terminal for travelers, while District 2 boasts commercial and cultural centers alongside a sprawling tourist green space. The outer boundaries of Districts 2 and 3 extend into non-residential highland areas.
• Region 9 comprises three districts. District 1 is home to the city’s largest university campus, with its southern sector bordering highland terrain. Districts 2 and 3 are emerging development zones, with their southern edges adjacent to highlands that have spurred recent urban expansion.
• Region 10, a rapidly growing urban area, currently sustains a sizable population. District 1 features extensive green spaces and commercial centers, while District 2 aligns with the city’s developmental trajectory.
• Region 11 features a large recreational green space in its District 1, along with commercial centers. District 2 of this region is aligned with the city’s urban development trends.
• Region 12, an expanding zone, is undergoing active construction and infrastructure development.
• Finally, the “Samen region” represents the city’s historic and religious core, hosting a concentration of hotels, traditional markets, and culturally significant landmarks.

2.2. Data Used

In this research, satellite images, statistical data and information, and organizational maps were prepared in different ways. To obtain the amount of waste production and per capita green space, the statistical book of Mashhad City in 2021 was used. The 2016 census data of “Mashhad Municipality Fava organization” was prepared as a geodatabase file using urban blocks and socioeconomic information for each urban block (data will be provided after submitting a request. Moreover, to use and identify the location of educational, medical, and public transportation land uses, the Shape file of “Mashhad Municipality FAVA Organization 2019” was used. To investigate traffic and transportation challenges as well as the actions of corporations, nitrogen dioxide (NO₂) and sulfur dioxide (SO₂) data derived from Sentinel-5p satellites [27] were retrieved from the Google Earth Engine (GEE) [28] platform. The datasets mentioned above were extracted from the average of two distinct periods, namely (22 September 2022, 21 March 2022) and (20 March 2022, 23 September 2021). To generate cartographic representations of land surface temperature (LST) and vegetation, Landsat 8 collection 2 level 2 (C2L2) satellite imagery was utilized. Given that the levels of greenness and temperature during June, July, and August exceed those of the other months, satellite imagery captured during these months was selected for analysis. The geodatabase files of Mashhad municipality were used to generate the maps of flood channels, faults, and land use areas. The STRM1 Arc-Second Global digital elevation model (DEM) map was obtained from the United States Geological Survey (USGS) website to generate DEM and slope maps. Furthermore, data regarding the volume of waste generation were extracted from the statistical compendium of Mashhad City. Additionally, the cartographic representation of the various regions and districts within Mashhad was derived from the geographic information system (GIS) dataset provided by the ministry of information and communications technology (ICT) [29].

2.3. Methodology

This study utilized GIS data, satellite images, products, and the 2021 statistical yearbook to generate spatial QOL indicators (Figure 2). The satellite images and products were obtained from two sources: USGS and GEE. Landsat 8 images were used to produce the normalized difference vegetation index (NDVI) and land surface temperature (LST) indicators. DEM was also obtained from USGS to produce the slope features. The SO₂ and NO₂ products were acquired from the Sentinel-5P satellite. After data collection, processing tasks such as averaging, spatial cropping, and resampling were performed. To integrate the indicators, it is necessary to map their numbers to the same scale. To this end, data normalization and reclassification were applied to standardize values to a common scale. Then, using a hierarchical structure and expert-derived weights, the FAHP method was used to generate an initial QOL map. Afterward, SVM produced the final QOL map using indicators and training samples from the initial QOL map. Finally, an accuracy assessment was conducted.

2.4. The Selected Indicators

The indicator selection process prioritized geospatial data (e.g., remote sensing-derived indices such as NDVI, LST, and air pollutants) and publicly available annual statistics (e.g., census data and socioeconomic indices) to ensure spatial granularity, temporal consistency, and alignment with established frameworks. These selections were informed by prior studies demonstrating their efficacy in urban QOL assessments, achieving an optimal balance between empirical rigor and operational feasibility.

Building upon the existing literature and considering Mashhad’s environmental context, we developed a comprehensive framework of criteria, sub-criteria, and indicators. While quality-of-life assessments typically incorporate both perceptual and objective dimensions, this study focused exclusively on spatial indicators. Subjective perceptions and mental aspects were excluded to maintain consistency with our geospatial analytical approach.

The selected indicators have considered all aspects from the environmental, biological, urban land use, and economic perspectives. Socioeconomic criteria and sub-criteria such as literacy level, employment, and population density have been included.

The criterion of urban land use has two sub-criteria: “access” and “per capita” urban land use. Therefore, in this research, important per capita urban land uses, including medical, educational, transportation, green space, and residential use, are considered [9,21,30]. Due to rapid population growth and urban expansion, traffic and environmental problems have emerged that affect the quality of urban life [31,32]. Moreover, land cover can also affect the QOL due to its role in reducing the temperature and pollutants [9,33]. Surface temperature varies seasonally, and urban heat islands—reflecting regional temperature differences—significantly impact QOL [21,25,33]. Another issue affecting the city’s pollution and QOL is the amount of waste production [10,34,35]. The natural environment is also a key component with two sub-criteria: natural hazards and natural features. Therefore, the indicators that can be effective in this field include fault, flood channel, elevation, and slope [36]. Hence, four main components affecting QOL, including 1—socioeconomic level, 2—urban land use (accessibility and per capita), 3—urban environment, and 4—natural environment, were considered. In Figure 3, the components, subcomponents, and indicators of the urban QOL model are presented.

2.5. Classification of Indicators

To integrate the maps, their values were standardized to a defined range. For this purpose, the “Reclassify” function in the GIS environment was used. All maps were reclassified on a scale of 1 to 5, where 1 indicates very poor QOL and 5 indicates very good QOL.

Table 1. Classification of urban quality-of-life indicators.

	Index	Unit	Very Good = 5	Good = 4	Medium = 3	Weak = 2	Very Weak = 1	Calculation Method	Reference
Socioeconomic level	Population density	p/ha	0–50	50–70	70–90	90–110	>110	p/area	[35]
	Employment level	%	80–100	60–80	40–60	20–40	0–20	${\displaystyle \frac{\mathrm{Employed}}{\mathrm{Employed}+\mathrm{Unemployed}}}$
	Education level	%	90–100	80–90	60–80	40–60	0–40	${\displaystyle \frac{Literate}{Literate+illiterate}}$
Transportation access	Bus	m	0–150	150–240	240–300	300–450	>450	Euclidean distance Rules	[37]
	Subway	m	0–400	400–500	500–800	800–1000	>1000	Euclidean distance Rules	[38,39]
Educational access	Kindergarten	m	0–300	300–350	350–400	400–500	>500	Euclidean distance Rules	[40]
	Elementary	m	0–500	500–600	600–700	700–1000	>1000	Euclidean distance Rules
	High school	m	0–1000	1000–1300	1300–1600	1600–2000	>2000	Euclidean distance Rules
	University	m	0–1000	1000–1300	1300–1600	1600–2000	>2000	Euclidean distance Rules
Access to treatment	Hospital	m	0–1000	1000–1200	1200–1300	1300–1500	>1500	Euclidean distance Rules	[41]
	Clinic	m	0–650	650–680	680–710	710–750	>750	Euclidean distance Rules
	Health center	m	500–0	500–530	530–560	560–600	>600	Euclidean distance Rules
Access to green space	Green space	m	0–400	400–600	600–750	750–1000	>1000	Euclidean distance Rules	[42]
Per capita urban use	Treatment	m2/p	>1.5	1–1.5	0.75–1	0.5–0.75	0–0.5	Euclidean distance Rules	[35]
	Educational	m2/p	>5	3–5	2–3	1–2	0–0.9	Euclidean distance Rules
	Transportation network	m2/p	>30	25–30	20–25	15–20	0–15	Euclidean distance Rules
	Green space	m2/p	>15	12–15	9–12	6–9	0–6	Euclidean distance Rules
	Residential	m2/p	>50	40–50	30–40	20–30	0–20	Euclidean distance Rules	[35]
Environment	NO2	ppm	0.108–0.141	0.141–0.172	0.172–0.207	0.207–0.239	0.239–0.27	Google Earth Engine	[35]
	SO2	ppm	0.178–0.225	0.225–0.278	0.278–0.332	0.332–0.383	0.383–0.434	Google Earth engine	[35]
	LST	C	29–36	36–42	42–48	48–54	54–55	Landsat8 C2L2	Global Warming and Changing the Range of Seasonal Temperatures
	NDVI		0.26–0.52	0.17–0.26	0.10–0.17	0.05–0.10	−0.12–0.05	Landsat8 C2L2	[43]
	Trash produced	Ton/ha/year	0–9	9–17	17–27	27–37	>37	Statistics	[35]
Natural environment	Fault	m	>1200	900–1200	600–900	300–600	0–300	Euclidean distance Rules	[44]
	Flood channel	m	>777	545–777	345–545	160–345	0–160	Euclidean distance Rules	[45]
	DEM	m	915–1000	1000–1100	1100–1200	1200–1300	1300–1341	DEM SRTM 1Arc-Second Global	[46]
	Slope	%	0–2%	2–5%	5–8%	8–12%	>12%	Spatial analyst/Surface/Slope	[47]

presents the classification of urban QOL.

2.6. Fuzzy AHP

In this research, Fuzzy AHP (one of the MCDM methods) was used, which uses fuzzy numbers to quantify expert decisions when comparing indicators. Like the analytical hierarchy process (AHP), Fuzzy AHP can provide weights for each indicator. In Figure 2, the structure of the research for preparing the QOL map is shown [48,49]. After preparing the initial QOL map, its values were classified into five classes: very weak, weak, medium, good, and very good. Pixels from each class were selected as ground truth data for applying the support vector machine classifier. The ground truth samples were grouped into training and test samples. The training samples were used to train the SVM classifier. Moreover, the test samples are utilized to assess the performance of the SVM classifier.

2.7. Support Vector Machine

Support vector machine (SVM) is a machine learning method that can be used as a classifier to group samples. SVM uses a decision surface to separate classes. Different decision boundaries such as linear, polynomial, radial basis function (RBF), and sigmoid kernels can be employed. The SVM approach was applied using some training samples. The SVM approach maximizes the distance between support vectors of classes. To perform the SVM classifier with RBF kernels, measuring its performance regarding changes in gamma and penalty parameters is necessary. To this end, gamma values of [0.01, 0.05, 0.1, 0.5, 1, 4, 7, 10] and penalty values of [50, 60, 70, 80, 90, 100] were selected [50].

2.8. Validation

Some samples obtained from the initial QOL map were used as test samples to validate the model. Classes estimated from the SVM classifier over the test pixels were compared with the ground truth data. A confusion matrix was generated to evaluate classification performance. After generating the confusion matrix, some statistical measures such as overall accuracy and Kappa coefficient were used to assess the effectiveness of the SVM method [48,49].

3. Results

3.1. Environmental Indicators

The model’s environmental indicators, including pollutants, waste production, and land surface temperature, were categorized according to [43]. For these three indicators, lower values indicate better conditions. Figure 4 depicts the result of the classification of environmental indicators. Figure 4a shows the land surface temperature in the city’s developing regions and the city’s southern outskirts, which have a higher temperature due to less construction. Figure 4b shows waste production in the areas. Figure 4c displays dense vegetation in park-covered areas. According to Figure 4d, NO₂ pollutants, primarily emitted from fossil fuel combustion, were concentrated in high-traffic zones and developing areas. Based on Figure 4e, SO₂ pollutants, a byproduct of industrial pollution, were found in the industrial areas with one or more factories.

3.2. Accessibility

Accessibility to urban land use was measured using a Euclidean distance analysis. Shorter distances to a land use correlate with QOL. Figure 5 presents the accessibility patterns for various urban land uses. According to Figure 5a–j, proximity to specific land uses shows a direct positive relationship with QOL, while greater distances correspond to reduced quality in terms of accessibility.

3.3. Natural Environment Indicators

Flood channels and faults were selected as key indicators of the natural environment. In this study, greater distances from the flood channels and faults were associated with better QOL scores. Moreover, based on the topography of the study area, lower slope gradients and elevations showed positive correlations with QOL. Figure 6b,d shows a low score near the flood channels and faults. Furthermore, the southern regions of the city demonstrated reduced QOL levels due to their higher elevations and mountainous terrain, as shown in Figure 6a,c.

3.4. Socioeconomic Indicators

The levels of employment and literacy were considered socioeconomic indicators for QOL. Lower scores were assigned to poorer QOL scores. Figure 7 depicts the spatial distribution of socioeconomic indicators across the study area. According to Figure 7a, the literacy level in Regions 4, 5, and 6 and the Samen region was slightly lower compared to that of the other areas. The percentage of employed people living in all city regions was higher than that of unemployed people (Figure 7b). The higher population density in Regions 1-6 and 10-11 correlated with reduced QOL scores (Figure 7c).

3.5. Per Capita Land Use

Figure 8 presents the per capita distribution of urban land uses. Figure 8a shows less residential land use per capita in the northern regions. Figure 8b,c depicts low educational and therapeutic land use per capita in most areas and districts. Figure 8d appears particularly limited in northern sectors. Figure 8e displays a lower transportation network per capita in Region 3.

3.6. FAHP Method

To perform the FAHP method, the weights of the indicators obtained from expert evaluations at multiple levels were calculated according to Figure 9. As can be seen, socioeconomic indicators emerged as the most effective in this study. Moreover, urban land use and environmental indicators both carried significant weights, approximately 0.2. In the socioeconomic sub-indicators, population (weight = 0.627) was identified as the most essential factor.

Using the weights of the FAHP method, indicators were generated using sub-indicators and their respective weights. The produced indicators, including air pollutants, accessibility and per capita urban use land, natural features, and natural hazards, are shown in Figure 10a–e. This figure depicts QOL statuses at five levels, ranging from “very poor” to “very good”. From the perspective of air pollution, blue regions in the east of the city demonstrated superior QOL conditions.

Moreover, regions in the center and north of the city exhibited very low QOL. From the viewpoint of access, the historic city center demonstrated superior access matrices compared to peripheral areas. According to the per capita per use map, the eastern areas were not in good condition, but the central regions of the city had appropriate QOL scores. Furthermore, blue and red areas of the natural environment and natural hazards maps depict strong and weak QOL, respectively.

3.7. Integrating Indicators of the Components

Figure 11a was obtained by integrating three indicators. According to the census information in the urban blocks, the quality score was high in Districts 0202, 0602, 0901, and 0903; Regions 7, 8, and 12; and the Samen region. The QOL of other regions was moderate to weak. This was because population density in urban districts had a higher weight, so the districts with higher population densities had lower quality. According to Figure 11b, the components of urban land use were obtained. The results in Regions 2, 3, and 4 and Districts 0602, 0603, 0503, 1201, and 1002 showed lower quality scores. Figure 11c includes urban environmental indicators such as air pollutants, ground temperature, vegetation density, and waste production. The results showed that QOL in marginal regions such as Regions 4, 5, and 12, as well as the Samen region was higher than in Regions 2, 9, 10, and 11. According to Figure 11d, the indicator of the natural environment is the sum of the slope, DEM, fault, and flood channel indicators. The regions with lower quality regarding the natural environment indicator include Region 9 and some parts of Regions 3, 4, 7, 8, and 11.

3.8. Analysis of Hot and Cold Spots

In this section, a hot and cold analysis was performed according to Figure 12. The blue cold spots represent lower-quality areas, while the red hot spots denote higher-quality zones. According to Figure 12a, the cold spots in Regions 3, 4, and 5 demonstrated lower quality status in the per capita urban land use index. As mentioned, the districts of the regions located on the edge of the boundary of the area have less accessibility than the central regions and districts. Therefore, according to Figure 12b, accessibility in the central regions was more than that in the boundary areas. Figure 12c shows that the regions with higher literacy rates appeared as hot spots. Moreover, cold spots were observed in regions 3–6 and the Samen region due to lower literacy rates. Figure 12d shows the regions with fewer employed residents, which were identified as cold spots. According to Figure 12e, hazard-free zones appeared as hot spots. A scattered distribution of hot and cold spots was observed according to Figure 12f, but there were more cold spots in Regions 8 and 9, indicating lower QOL for this indicator.

According to Figure 12g, Region 12 had become a blue spot with a higher temperature due to less construction of the Earth’s surface. Also, according to Figure 12h, the state of vegetation is shown as hot spots in the parts where the density of natural cover was higher. In Figure 12i, the status of waste production is shown as a hot spot in the Samen region due to less waste production among other areas. The cold spots in Figure 12j indicate lower quality due to more pollutants.

3.9. QOL with SVM

The SVM method successfully generated a QOL map from the indicators and training samples. The results obtained from this method are presented below. A comprehensive sensitivity analysis was conducted for the gamma and penalty parameters. Figure 13 shows the overall accuracy of the test sample for different gamma and penalty values. It was concluded that gamma was the most effective parameter for the accuracy of results. The optimal configuration (Penalty=0.5) achieved 98% overall accuracy—the highest among all experiments. The effect of the two parameters on the kappa coefficient is presented in Figure 14. Consistent with the overall accuracy, a penalty value of 0.5 yielded superior performance. Moreover, the gamma parameter showed minimal impact on the results.

The confusion matrix of the best experiment is presented in Figure 15. The matrix includes five QOL classes: 1—very weak (VW), 2—weak (W), 3—medium (M), 4—good (G), and 5—very good (VG). According to this table, 42,167 VW pixels were correctly classified as VW, and 534 VW pixels were misclassified as W. Accuracy in the detection classes was about 97 to 98%, which indicated that the SVM method was robust for creating a QOL map. The output of the SVM method, i.e., the QOL map, is presented in Figure 16. According to the visual analysis, the output of the SVM method was similar to the FAHP QOL map.

4. Discussion

The total QOL of Mashhad City is presented in Figure 17. As shown, Districts 1201 and 1202, were in a better condition in terms of the socioeconomic, environmental, natural environment, per capita, and accessibility indicators. Employment and literacy scored higher, while population density obtained lower QOL scores in these districts. These districts also had more potential to improve transportation and residential per capita for the city’s future development. Natural hazards, such as faults, landslides, and natural features like the land’s topography, had suitable conditions in these districts. The southern and southeastern regions, including Regions 9, 8, and 7 and the Samen region, showed two different levels of QOL scores. Districts 0902 and 0903 had lower qualities than Regions 8 and 7, and the Samen region due to their altitudes, pollutants, faults and channels, and higher population density. Districts 0803 and 0703, and the Samen region had better conditions due to better heights and topography, fewer pollutants, and lower population density. Although the districts of Regions 2, 3, 4, 5, and 6 were in good condition regarding environmental parameters and natural environment, they had lower quality regarding population density, per capita, and accessibility of urban land uses. The QOL map of District 0202 showed better conditions compared to the neighboring districts since indicators such as their population density, transportation per capita, and distance from natural hazards were better. Also, Regions 10 and 11 had poor quality in terms of population density, per capita urban land use, natural environment, and ecological environment due to their fault lines and air pollutants. Region 1 was in a low-quality state only regarding air pollutants. However, per capita, the accessibility of urban land uses, and the natural environment showed that it has good QOL.

In general, the QOL in the 13 regions of each district according to all indicators and criteria is presented in

Table 2. The percentage of pixels in each class of QOL map in the districts of each region.

Regions	Districts	District’s Code	Very Weak	Weak	Medium	Good	Very Good	Quality
1	1	0101	-	0.4%	99.5%	-	-	3
	2	0102	0.05%	90.6%	9.2%	-	-	2
2	1	0201	98%	1.4%	-	-	-	1
	2	0202	-	-	2.2%	97.7%	-	4
	3	0203	100%	-	-	-	-	1
	4	0204	100%	-	-	-	-	1
3	1	0301	85.9%	14%	-	-	-	1
	2	0302	100%	-	-	-	-	1
	3	0303	39%	60%	-	-	-	2
4	1	0401	26%	73%	-	-	-	2
	2	0402	86%	13%	-	-	-	1
	3	0403	89%	10%	-	-	-	1
5	1	0501	-	0.4%	99%	-	-	3
	2	0502	-	2.9%	97%	-	-	3
	3	0503	6.2%	93%	-	-	-	2
6	1	0601	0.2%	98%	1.3%	-	-	2
	2	0602	-	4.5%	95%	-	-	3
	3	0603	97.8%	2.1%	-	-	-	1
7	1	0701	-	-	-	59%	40%	4
	2	0702	-	1.9%	98%	-	-	3
	3	0703	-	-	-	46%	53.9%	5
8	1	0801	-	-	-	-	100%	5
	2	0802	-	-	-	16%	83.9%	5
	3	0803	-	-	-	93.9%	6%	4
9	1	0901	-	-	0.7%	99%	-	4
	2	0902	4.3%	95%	0.6%	-	-	2
	3	0903	0.01%	26%	73%	-	-	3
10	1	1001	0.5%	82.9%	16.4%	-	-	2
	2	1002	99.7%	0.2%	-	-	-	1
	3	1003	87%	12.9%	-	-	-	1
11	1	1101	32.7%	67%	-	-	-	2
	2	1102	0.4%	84%	15%	-	-	2
12	1	1201	-	-	-	97%	2.7%	4
	2	1202	-	-	-	7.3%	92.6%	5
Samen			-	-	-	6.2%	93.7%	5
			-	-	-	0.15%	99.8%	5

. The QOL map was classified into five classes. Statistics were estimated based on pixels within each district. The quality of District 0101 was classified as class 3, medium, since about 99% of pixels within this district were classified as medium. Moreover, District 0302 was classified as class 1, very weak, because 100% of pixels of this district were classified as “very weak”.

The results of this study were compared with similar studies conducted in Mashhad City [18,21]. The proposed method showed that Regions 8, 12, and 7 and the Samen region had good QOL, consistent with the results of [50], which identified Regions 8, 12, and 7 as medium to developed areas. Moreover, a study using the TOPSIS method showed that Regions 12 and 7 had high QOL [18]. Another study using the FANP method showed that quality decreases toward the marginal parts of the city, confirming the results of this study [21].

The final quality-of-life results for each indicator in all regions and their districts of Mashhad City are presented in

Table 3. Final status of quality-of-life indicators in each district of Mashhad City.

Regions	Districts	Socioeconomic Level			Urban Land Use		Environmental				Natural Environment		QOL
		Population Density	Employment	Literacy	Access	Per Capita	Waste Production	Air Pollutants	LST	NDVI	Natural Feature	Natural Hazards
1	1	2	5	5	3	4	3	1	4	2	4	5	3
	2	1	5	5	5	3	3	1	4	2	4	5	2
2	1	1	5	5	4	2	1	1	4	2	4	5	1
	2	5	5	5	1	3	1	2	3	1	4	5	4
	3	1	5	5	1	2	1	2	3	1	4	5	1
	4	1	5	5	2	2	1	1	4	2	4	5	1
3	1	1	5	5	4	2	1	2	4	2	4	5	1
	2	1	5	5	1	1	1	5	3	1	4	3	1
	3	2	5	5	1	1	1	5	3	2	4	5	2
4	1	1	5	4	5	2	2	3	4	2	4	5	2
	2	1	5	4	4	1	2	4	4	2	4	5	1
	3	1	5	4	1	1	2	5	4	1	4	3	1
5	1	2	5	5	2	3	3	3	3	2	4	5	3
	2	2	5	4	1	3	3	4	3	1	4	5	3
	3	1	5	4	4	1	3	5	3	1	4	3	2
6	1	1	5	5	2	3	2	3	4	2	4	5	2
	2	3	5	4	1	2	2	4	3	1	4	5	3
	3	1	5	4	2	1	2	4	4	1	4	5	1
7	1	5	5	5	1	3	2	3	3	2	4	5	4
	2	3	5	5	1	3	2	1	3	2	4	3	3
	3	5	5	4	1	5	2	4	3	1	4	5	5
8	1	5	5	5	5	5	3	1	4	2	4	5	5
	2	5	5	5	1	5	3	1	3	2	1	3	5
	3	4	5	5	1	4	3	1	4	2	4	3	4
9	1	5	5	5	1	4	1	2	4	2	1	3	4
	2	2	5	5	1	4	1	1	4	2	1	2	2
	3	3	5	5	1	4	1	1	4	2	4	3	3
10	1	2	5	5	2	4	1	1	3	1	4	5	2
	2	1	5	5	4	2	1	1	3	2	4	5	1
	3	1	5	5	2	3	1	1	3	1	4	5	1
11	1	1	5	5	4	3	2	1	4	2	4	5	2
	2	2	5	5	2	3	2	1	4	2	4	4	2
12	1	5	5	5	1	3	4	1	2	1	4	5	4
	2	5	5	5	1	5	4	1	2	1	4	5	5
Samen	N/A	5	5	4	4	5	5	2	4	1	4	5	5

. For example, District 0801 received a score of 5, indicating very good QOL, on average. The analysis of indicators showed very good QOL in terms of population density, employment, literacy, accessibility, per capita, and natural hazards. Moreover, District 0203 had a very poor quality of life, as its population density, urban land use, and environmental indicators were not in good condition. Furthermore, District 0501 had a medium quality of life due to the impact of urban land use and environmental indicators on the final results.

Given that in most districts with lower QOL, the main problems and challenges revolve around rising air pollutants, waste production, lack of green space, limited access to urban land uses, and high urban population density, urban managers and planners can improve the quality of life in Mashhad City by considering the following strategies:

For air pollution mitigation, expand green public transportation by developing metro lines, electric buses, and bicycle lanes to reduce vehicle emissions; establish stricter industrial emission standards; and monitor factories while enforcing advanced air filtration technologies. Plant trees and create green belts to absorb particulate matter and improve air quality, especially in densely populated areas. For waste management, implement waste segregation at the source, and educate citizens to reduce their per capita waste production. To expand green spaces, transform vacant lots into local parks; ensure equitable distribution of green spaces in underserved areas; develop green roofs and vertical gardens, particularly in high-density central zones; protect existing natural areas; and preserve forests and wetlands on the city’s outskirts. For urban land use optimization, promote mixed-use development; integrate residential, commercial, and service facilities to reduce urban travel needs (mixed land use); revitalize decaying urban areas; convert abandoned spaces into cultural, sports, or educational hubs; decentralize policies; and prevent overconcentration of services in central districts. For population density control, create new residential hubs around Mashhad to absorb excess population; upgrade infrastructure in dense areas; expand water, electricity, and sewage networks to meet demand; limit construction permits in saturated zones; and focus on urban renewal instead of new developments to help reduce the population density in the city of Mashhad.

5. Conclusions

This paper provides a comprehensive analysis of QOL in Mashhad City. The integration of FAHP and SVM methods along with the proposed QOL framework represent the novel aspects of this study. Using training samples extracted from the FAHP-derived QOL map for SVM training was another innovation of this work. Moreover, 4 main components, 11 subcomponents, and 17 indicators derived from remote sensing and GIS data were introduced. The results showed that remotely sensed indices such as LST, NDVI, and air pollution significantly enhance QOL assessment accuracy in the study area. The FAHP-weighted QOL map revealed that high-density regions had poorer quality of life. Moreover, from the perspective of the urban environment and air pollution, areas on the margin of Mashhad City demonstrated better urban environmental conditions and air quality. The results of the proposed method showed that Regions 8, 7, and 12, and the Samen region had good QOL scores; Regions 1, 5, and 9 had moderate QOL scores; and Regions 2, 3, 4, 6, 10, and 11 had weak QOL scores. The primary challenges in the low-QOL districts included the air pollutants, waste production, lack of green space, limited access to urban land uses, and high urban population density. Given Mashhad’s expansion to 17 regions, ongoing QOL monitoring using updated demographic data and environmental metrics is essential.

The results of the SVM classifier were more accurate in producing a QOL map.

A parameter sensitivity analysis confirmed gamma and penalty as crucial for SVM performance. SVM outperformed the other methods in QOL mapping, with a 97.0 kappa coefficient and 98% overall accuracy. It showed that the proposed method was robust enough for this study.

In order to investigate the quality of life in cities, the integration of other methods with machine learning algorithms such as FAHP+ ANN and FAHP+ Random Forest is suggested. It is also possible to use more recent satellite images with higher radiometric and spatial resolution or examine the environmental parameters separately during the year’s seasons and evaluate the changes in the QOL of urban zones. In addition, it is suggested that the proposed method be implemented in more cities to create a suitable database from its output. This could lead to models that can be implemented in other cities in the future. It is suggested that in future studies, citizens’ views on the quality of life of their place of residence should be prepared in a location-based and questionnaire-based manner to be used as input for the machine learning method, which also includes perceptual indicators.