AHP and Machine Learning-Based Military Strategic Site Selection: A Case Study of Adea District East Shewa Zone, Ethiopia

,


Introduction
Geospatial artificial intelligence (GeoAI) is a developing scientific discipline that combines innovations in spatial science, artificial intelligence methods in machine learning (deep learning), data mining, and high-performance computing to excerpt knowledge from spatial big data [1,2]. Site selection is an essential component of military strategy [3].
Suitability analysis is the process used to establish the suitability of a system according to existing geospatial data [4][5][6]. Remote sensing, geographic information system (GIS), analytic hierarchy process (AHP), and expert system (ES) methods are vital tools for identification, and comparison and multicriterion decision-making (MCDM) analyses are the best models for selecting military strategic sites [7][8][9][10]. The concept of the hierarchical analysis process was coined by [11]. It is used to solve complex decisionmaking processes and help the decision-maker make the best decision. It reduces the complex decision to a series of pairwise comparisons to produce results [7,12]. The AHP can be a powerful tool for addressing problems with multiple interrelated goals [13]. AHP is one of the most commonly used mathematical techniques for building hierarchical models for defining multicriteria decision-making (MCDM) [14]. Machine learning has gained the public's attention as a beneficial tool for evaluating site selection criteria and delivering simple messages and information from massive amounts of data [15]. The machine learning (ML) model by supporting an expert system will calculate the weights of different factors and compute the probability of the best-selected site [16][17][18]. Gradient descent was used for weight redistribution in states when indicators are missing to better predict the decision once all the indicators are known [19]. Gradient descent is a machine learning model training optimization algorithm. Based on historical data, the adaptive gradient method proposes that the learning rate for each parameter be adjusted during the learning phase [20][21][22]. Any machine learning methods are not necessary for the filter method [23]. Each variable is given a score and evaluated using statistical methods. Machine learning models are used for selecting military strategic site selection by using the geospatial data classification method [24]. The variables are then ranked based on the score. Filter methods have the benefits of being quick, scalable, model-free, and having minimal computational complexity [25,26]. In order to train the ML models, a set of satellite image-classified data is required [27,28].
Machine learning tools that can be effectively combined to create intelligent systems for war planning include geographic information systems (GIS), artificial intelligence, and remote sensing [29,30]. Because ML techniques use computational algorithms and statistics to develop a new model, they produce more accurate results [31][32][33][34][35]. Over the last few years, the geographic information system has been used for the operation and representation of geospatial data in suitability analysis [36][37][38]. These technologies can be employed effectively in systems that involve command, control, communication, coordination, and information. Satellite remote sensing data can be used to create a variety of outputs, including land use/land cover maps, obstacle maps, slope maps, road mobility maps, and line of sight plots [39][40][41]. Utilizing GIS, machine learning, and RSbased techniques allowed us to choose better locations for military strategy and border control in order to identify potentially hostile environments [42,43]. These techniques can also produce more complete data and information on which to base a site selection decision [44].
Site selection using GIS, remote sensing, analytical hierarchy process, and machine learning is a well-established use of geospatial data [14,[45][46][47][48][49][50][51]. Armed forces use GIS in cartography, combat area management, military deployment, intelligence, and terrain analysis, allowing decision-makers and commanders to obtain useful data and information [52][53][54]. GIS, analytical hierarchy method, and machine learning are commonly used tools that allow users to select the best strategic sites among numerous criteria [55]. The integration of a geographic information system and multicriteria evaluation approaches helps decision-making by creating an environment conducive to managing and organizing large amounts of geographical data [56]. GIS performs deterministic overlay and buffer operations in site selection problems, while multicriteria techniques analyze alternatives based on the decision-maker's subjective values and priorities [55]. Multicriteria decision systems, when combined with GIS, can help decision-makers choose the best site [57,58]. The combination of GIS and MCE is a powerful tool for choosing strategic military locations and producing top-notch analytical results. The main criteria and data used for military strategic site selection in Ethiopia are existing all facilities, rivers, roads, land use/land cover, soil types, settlements, DEM data, satellite imageries, drainage data, geology, tactical consideration data collected by GPS, obstacles, peak area, and slope data [59][60][61][62]. The aim of the proposed study is to select and propose new strategic military suitable sites by considering international standard military site selection factors and criteria to test and evaluate selection criteria by using an analytical hierarchy and machine learning algorithm compared with actual camp locations or vacated sites to determine optimal sites for the military base.

Study Area. Adea District is located in the East Shewa
Zone of the Oromia Region at a distance of 48 km from Addis Ababa in the Great Rift Valley. The district geographically lies within 8°38 ′ 47 ″ to 8°56 ′ 52 ″ N latitude and 38°53 ′ 30 ″ -39°11 ′ 24 ″ E longitude. The area coverage of the district is 936.41 square kilometers. According to the 2007 national census, the total population of the woreda is 130,321, of whom 67,869 were men and 62,452 were women. Adea has a high Great Rift Valley the elevation of the area varies from 1,680 to 2,867 m almost surely. According to the National Meteorological Services Agency, Adea District gets a maximum rainfall amount of 1,716 mm and a minimum of 1,220 mm annually. The highest mean maximum temperatures of the woreda were about 28°C from February to May and 24°C from September to November, and the lowest mean temperatures were 23°C or lower between June and August ( Figure 1).

Data and Software.
To suggest suitable sites for strategic military bases, the analytic hierarchy process (AHP) [11,63] method was used. Geographic information system-(GIS-) based multicriteria decision-making technique proves to be efficient in these conditions owing to its ability to manage large volumes of spatial data from various sources [64]. For GIS-based analysis, Esri ArcGIS 10.8, Arcpro2.9, Erdas Imagine 2015, ENVI 5.0, machine learning algorithm, Jupyter Notebook, and AHP-Idrisi-based software are used in this study. The data used for analysis can be broadly classified into two groups. Primary data includes raw data obtained from different sources like the USGS Earth Explorer and government and nongovernment institutions. Journal of Sensors Secondary data includes the processed data obtained from primary data and analytical hierarchy weighted-based analysis in machine learning operations ( Figure 2 [65] strategic military site selection document and considered proximity to access to roads, urban areas, airports, and the proximity to railway stations to protect civilians from chemicals used by armed force (Figure 3).

Factors Considered for Evaluation Comparison Class.
This study adopted several datasets including raster as well as vector for deriving different information for determining military strategic suitable site selection. The present study also used important physical parameters which are considered for the suitability analysis: slope, altitude, land use/ land cover, geology, soil, population data, and existing shapefiles like roads, rivers, lake settlements, and airports. The datasets, their sources, and the purposes for which they have been used are summarized in  are then given the weights of the different criteria and aggregated using the weighted linear combination (WLC) based on MCDM and the GIS tool of the weightage as per Saaty's AHP (analytic hierarchy process) method [11] by creating a separate field in each layer in Arc GIS 10.8, Erdas Imagine 2015, and machine learning algorithms. Each class associated with each layer is given a rank and stored as a separate field in the geodatabase. The product of weightage and rank is computed and stored in another field. These vector layers are then integrated into a GIS environment. Under the AHP method, problems are modeled by creating a hierarchical structure representing the determined alternatives and criteria (Table 2). MCDM was adapted to determine arbitrary and conflicting requirements by decision-makers and research hierarchy processes. This method is widely used for aggregating GIS [66]. The relative importance between the two criteria is measured on a scale of 1 to 9 given by Saaty. The weight of each criterion is calculated based on a pairwise comparison of different criteria [55]. AHP also provides an efficient method for examining the consistency of the findings, which lessens bias in the decision-making process. The consistency test is performed after the normalized weights of the criteria, which are defined by the level of importance. The consistency index (CI) and consistency ratio (CR) values are determined using the equations of [67,68]. The extent of transitive consistency that can be maintained is the key to stratification analysis, and this can be validated by determining the consistency ratio (CR), which divides the consistency index (CI) of the result using a random index (RI) Table 3. The formula for CI is CI = λ max − n/ð−1Þ, λ max is the principal eigenvalue and n is number of factors. A mathematical equation was used to calculate it.
2.6. Analytic Hierarchy Process. In the analytic hierarchy process (AHP), a rater makes pairwise comparisons between alternatives. We collect these comparisons into a matrix. Ideally, the ij-th element in the matrix is the ratio of the priorities of wi and wj of the i-th and j-th alternatives. The square matrix aij is an estimate of the relative weight wi/wj of i for element j, and matrix A is aij = 1/aji, a reciprocal matrix where the elements in the main diagonal are all equal to 1. If we have N alternatives, then the matrix looks like Table 4.
The smaller the value of CR, the higher its transitive consistency. Saaty judged that people performed a two-way comparison fairly consistently if the CR was within 10% (0.1), and  5 Journal of Sensors an acceptable level of inconsistency was within 20% (0.2); however, a lack of consistency above 20% (0.2) required reexamination. In this study, only CR results within 0.1 were utilized. In CR, the consistency ratio varies from 0 to 0.178; CI is the estimated consistency index according to the equation.

Developing Machine Learning
Model. Machine learning methods have become a standard solution for processing big data analytics [69,70]. Classification accuracy is defined as the ratio of accurate predictions (true positive plus true negative) divided by the total number of predictions made (true positive plus true negative plus false positive plus false negative).

Accuracy = TP + TN TP + FP + TN + FN
: For each feature in the dataset, TP, TN, FP, and FN are first calculated, and then they are cumulated. The terms TP, TN, FP, and FN have the following definitions ( Figure 4).

Criteria for Site Analysis.
The optimal criteria for strategic military site selection are examined for doling out relative positions and individual factor weight reliant on the land use type for which wellness is being considered. For the importance of the investigation, the criteria under concern in this paper are well organized (Table 5).
2.9. Prioritization Criteria. The main prioritization criteria for selecting strategic military site activities are identified as follows (Table 6): AHP scale: 1 as equal importance, 3 as moderate importance, 5 as strong importance, 7 as very strong importance, and 9 as extreme importance (2, 4, 6, and 8 values in-between). These are the resulting weights for the criteria based on your pairwise comparisons.
The analytic hierarchy process (AHP) is a discrete measurement result that derives scales of values from pairwise comparisons and ratings. The above table shows the negative and positive values, which means that in decision-making, there are often criteria that are opposite in direction to other criteria, such as benefits versus costs and opportunities versus risks of site selection, and these must sometimes be distinguished by using negative and positive numbers. This means that the concept of consistency is critical to the AHP. The positive ideal solution minimizes the cost criteria while maximizing the benefit criteria, whereas the negative ideal solution maximizes the cost criteria while minimizing the benefit criteria. As a result, both negative and positive values are the same number in this study's result, indicating that our model scale for a site selection criterion is possessed by other criteria used to measure that suitability between them, the relative weights of those criteria with respect to that criteria are well balanced, and our model is accurate. One element is absolutely more important than the other The evidence in favour of one element over another is as convincing as possible A reasonable assumption

Reciprocals of above
If activity i has one of the above nonzero numbers assigned to it when compared with activity j, then j has the reciprocal value when compared with i if the activities are very close May be difficult to assign the best value, but when compared with other contrasting activities, the size of the small numbers would not be too noticeable, yet they can still indicate the relative importance of the activities. 2, 4, 6, 8 Intermediate values In CR, the consistency ratio varies from 0 to 0.178; CI is the estimated consistency index according to the equation. 6 Journal of Sensors 2.10. Accuracy Assessment. After classification, an accuracy assessment was used to determine the level of acceptance for the LULC maps that were produced. By comparing the error matrices with the reference data, it is usual practice to evaluate the classification accuracy of LULC types [71].

Results
The identification of a strategic military suitable site is very important to support Ethiopian military forces. The physical parameters affecting the strategic military site suitability selections are discussed below. The cumulative effect of these factors determines the degree of best strategic suitable military strategic site establishment and also helps in identifying the limitations of the current established military sites. The various map layers generated to serve the purpose are shown in Figure 5.
3.1. Land Use and Land Cover. The study area's land use/ land cover map was divided into six categories: shrubland, farmland, forest, grassland, built-up area, and waterbody. Farmland covers the majority of the study area (56%), followed by built-up area (28.1%) and grassland (7%). The central part of the study area is covered by agricultural/cropland and grassland, which are ideal for strategic military sites, based on a fuzzified land use/land cover map. With a membership value of 0, the area under built-up is fuzzified as unsuitable. In addition, land use/land cover in the area where the military strategic suitable site location is important, especially in terms of minimizing environmental and social impacts. Depending on land use/land cover data, the selected site for military strategic sites in Adea District is farmland, grassland, and shrubland areas that are appropriate for the construction of military camps (Table 7).
3.2. Geology. Geological phenomena are also another factor in site selection, and it includes a natural phenomenon involving the structure or composition of the earth. Identifying the geological features helps to study the suitability and stability of rock type for command post best site selection. The selected site for military strategic site geology type includes Middle Miocene and Pliocene-Pleistocene because it includes alkaline basalt and trachyte. Alkaline basalt is the best and most selected geology for military site selection because this type of rock is used for construction (i.e., building blocks or in the groundwork), making cobblestones (from columnar basalt), and making statues. Heating and extruding basalt yields stone wool, which has the potential to be an excellent thermal insulator.  3.5. Population. The strategic military site selection needs access but is far from very high population density, proximity to multimodal transportation, and interest from the community and government. Cities have always been an important factor in the power game, and in an armed conflict, the taking or destruction of them can become a symbol or an end in itself. But in principle, the military campsite is far at least 1 km from a densely populated area. Every military site must have an exclusion area, a low population, and site characteristics such as adequate security plans and a well-measured area. Geology Geological suitability (Quaternary) 9 9

Elevation and
Geological suitability (Middle Miocene) 7 6 Geological suitability (Pliocene, Pleistocene) 3 9 Geological suitability (alkaline basalt & trachyte)   But it must be far from cities and towns because of population safety. The site selected by using multicriteria analysis in terms of cities and towns is appropriate. Accessibility to main roads (primary, secondary, and tertiary roads only) is evaluated using the following principles: very poor access (>25 km), poor access (20-25 km>4/5 hrs), medium access (15-20 km3/4 hrs), good access (10-15 km-2 hrs), and very good access (10 km) 1 1/2 hrs.

Rainfall. Strategic site selection for military operations is
significantly impacted by the variety of climatic and extreme weather phenomena, such as high and low temperatures, droughts and floods, strong and destructive winds, and heavy or blowing snow. These effects include an increase in the risk to life and safety, harm, and a decline in mission effectiveness. Rainfall has an impact on the condition of the ground, visibility, the efficiency of employees, and the operation of some equipment. Heavy rain may make some unsurfaced roads and off-road areas impassible. Generally, precipitation in excess is not appropriate for strategic military site selection. In this case, the annual rainfall between 1,220 and 1,716 mm is used for selecting the best strategic military site. According to rainfall statistics, the average annual rainfall ranges from 1,280 mm to 1,421 mm, with a mean value of 1,300 mm. The most precipitation occurs between June and September, accounting for 75.23% of total annual precipitation, with the short rain season (March to May) accounting for 16.75% and dry months accounting for a meager 8.02%. Despite bimodal rainfall, short showers (Belgium) are erratic and inadequate for rainfed crops. Furthermore, in most parts of the study area, rainfall and military site selection have a high positive correlation, implying that rainfall influences the success and obstacles of military operations ( Figure 5 (a) Proximity to the airport, (b) proximity to a water resource, (c) proximity to access road and railway station, (d) elevation, (e) proximity to geological fault and geology type, (f) land use/land cover type, (g) soil type, (h) slope steepness and gradient, (i) average annual precipitation, (j) mountains and contour lines, (k) proximity to urban areas, and (l) spot height and topography are included in Table 8 (Figure 6).

Evaluation of Selected Criteria
3.9.1. Decision Matrix. The weight matrix with the design criteria and the rating for strategic military site consolidation results is shown in Figure 7. The analytic hierarchy process (AHP) is a technique used to derive ratio scales from paired comparisons that are used to organize and analyze complex decisions. In order to develop rankings of the suitable site for strategic military area consolidation in the research area, twelve factors describing the entire spatial structure of the surveyed district were designated. According to the results, elevation and road contribute the most; geology, soil, land use, slope, precipitation, and mountain area contribute mod-erately; and the other criteria classes contribute little to the selection of strategic military sites. Compute the ratios of their areas exactly and enter these ratios instead of using judgments; the priority vector will return the exact relative areas of the figures.
In the final stage, the relative importance of each criterion for strategic military site selection was assigned a weight. In order to establish a final suitable site, this was done by utilizing analytic hierarchy process tools to standardize the matrix that we used in the weighted overlay spatial analysis.
Based on the table and figure results, the number of comparisons is 66, principal eigenvalue is 13.193, consistency ratio CR is 7.1%, eigenvector solution is 6 iterations, and delta is 2.0E-8 ( Figure 7 and Table 10). N.B: Sl = slope, P = precipitation, G = geology, S = soil, LU = land use, E = elevation, W = waterbody, Mt = mountain, Bu = built-up area, Rd = roads and railways, Ap = airport, and Sh = spot height. Where Si is the suitability index, RF is the rainfall criterion, S is the soil criterion, G is the geology criterion, Mt is the mountain point criterion, El is the elevation criterion, SL is the slope criterion, BU is the residential area criterion, Rd is the road and railway criterion, R is the river criterion, LU is the land use criterion, Ar is the airport location criterion, and Sp is the spot height point criterion (Table 11).
3.9.2. Accuracy Assessment. In order to determine the producer, user, overall accuracy, and Kappa coefficient, this study used [71]. The Kappa coefficient of agreement was employed for multivariate statistical analysis, which could be used to determine the relationship between categorized samples and reference data [73]. The assessment accuracy of LULC maps for the year 2021 was 97.00%, and the Kappa coefficients were 0.96 (Table 12).

Model Validation of Potential
Sites. The accuracy measure can be used to determine the classification model's effectiveness. The accuracy of machine learning models can

Journal of Sensors
To specify (labeled) parameters, both dependent and independent variables between the criterion class and supervised machine learning models are developed. The decision tree (CART) is then taken in advance by the random forests with an assessment to make the employed trees more inde-pendent (less correlated). In order to use Python notebooks for the classification of the 1,449 instances, I have created a predetermined number of random punctual entities (1,449) at random points of Adea woreda's strategic military sites, removed the attributes from the criteria maps and AHP    Journal of Sensors maps, and then randomly selected the points for the study. Following this, we applied the random forest technique, which produced 1,449 instances and 6 attributes. A sizeable training sample is needed in order to achieve decent results because the numbers of entities in the two subgroups established at each binary split, which correspond to the two branches coming from each intermediate node, decrease over time. Proximity to land use, proximity to road and railway stations, elevation, slope, geology, soil, proximity to water resources (i.e., river and lake), precipitation, mountain and contour lines, spot height, proximity to settlement, proximity to airports, class AHP. Test mode: split 20.0% train, remainder test, and classifier model random forest of 12 trees, each constructed while considering 5 random features. Out-of-bag error: 0.0129 from the evaluation on test split ( Figure 11). R-squared is most frequently used to determine how well a regression model explains observed data. R-squared provides a straightforward 0-1 scale for evaluating the strength of the link between your model and the dependent variable.

Journal of Sensors
The result of the above figure showed the R 2 result of our model is 1 and it is the best fit because a greater R-squared typically means the model is better at explaining the variability. According to Wang et al. [74], the coefficient of determination is invariant for linear transformations of the distribution of the independent variables, and any output value that is close to 1 indicates a reliable prediction regardless of the scale on which these variables are assessed.

Discussion
By selecting a number of criteria and using satellite data, the current study addressed the demand for Adea District strategic military site selection. To address the issue, a technology based approach integrating a geographic information system (GIS), remote sensing, machine learning algorithm, and analytical hierarchy process (AHP) was used. This tool's ability to gather and manage a large volume of data makes it useful for site selection research, especially for strategic military site. In recent years, this method has been widely applied in site selection research. Using GIS and different multicriteria decision-making techniques, more than 100 studies on site suitability have been conducted, according to a recent study [75]. Many researchers all over the world have expressed an interest in using multicriteria design and Ahp techniques to select suitable sites [76][77][78][79]. In addition, the analytical hierarchy model takes into account environmental concerns such as hydrological, topographic and climatic criteria, land features, access, and infrastructure, and other relevant criteria were chosen from these to support the decision in selecting a suitable site [80]. As a result, many scholars and researchers have used MCDM techniques [42,81,82]. As such, the current study preferred to use a random forest and the fuzzy AHP ensemble with ML techniques to find a suitable strategic military site. In light of these kinds of findings, the present study also favoured using the fuzzy AHP ensemble with ML techniques to identify appropriate landfill sites for sustainable SVM [83].
In order to train a site selection model with better and more accurate performance, the ensemble learning idea is used in this study. According to research by [84] on the application analysis of the machine learning fusion model used to create a financial fraud prediction model, a single model may have flaws such as excessive bias or inadequate generalization. Also, Kahinda et al. [85] developed a GISbased military site suitability selection model for assessing the suitability of strategic military site in South Africa by merging physical, environmental, safety, and constraint layer using MCA. Another several studies [86][87][88][89][90] examined the parameters required for choosing prospective strategic military sites. Other researchers identified criterion layers for analysis that were not equally significant in directly choosing probable strategic military sites in the study area. However, using the AHP model allows you to overcome this difference.
Furthermore, based on classification and regression trees (CART), random forest (RF) models generate numerous independent trees to arrive at a final decision (e.g., majority votes for classification and average votes for regression). This is done by using two randomization approaches: the selection of training samples and the selection of variables at each node of a tree [91][92][93]]. An RF model randomly permutes each variable, and the regression mean square error (MSE) or classification error rate (CER) using out-of-bag (OOB) data is used to assess the relative relevance of the model variables [94,95]. In contrast to previous studies [7,[96][97][98][99][100][101], a greater number of subcriteria were employed for site selection [102,103], respectively. However, the number of subcriteria in this type of study is determined by the study's focus and the availability of spatial data. Finally, Gigovic et al. [104] suggest a comparison of the results with those produced from the most commonly used MCDM approaches, as well as a sensitivity analysis of the alternative rankings in relation to changes in the weight coefficients of the assessment criteria, in their study.

Conclusion
Limited available land in relation to prospective operational area can be overcome through site suitability research for military strategic sites. The GIS-based multicriteria evaluation technique is relatively straightforward and adaptable, and it may be used to assess the feasibility of a hilly area for various purposes. ML is one of the most exciting tools to have recently entered the geospatial science toolbox. Today, artificial intelligence methods, especially ML techniques, have come to the attention of scientists and officials in military science fields to analyze and manage the enormous amounts of data that are produced at any given time. GIS is unquestionably one of these fields. The GIS's data can be utilized for a variety of things, including military strategic site selection, transportation, drought analysis, agriculture, disease outbreak analysis, and land occupation. In addition, GIS makes it possible to store a large volume of data related to military purposes in a secure location and access it quickly whenever needed. Furthermore, by cutting down on the time needed for site selection, the combination of GIS with ML offers a potential method for lowering the cost of spatial information processing. Through integration, it is possible to transfer the interpretive findings from a small area to a larger, geographically similar area without having to spend extra time and money sending geographers out into the field for a long enough period of time to cover the entire area. Finally, this research will enable decision-making for choosing an appropriate military strategic site, and it also presents a chance to enhance site planning and give a management tool for military purposes.

Limitations
Site selection for the military is a difficult task, and using various machine learning and remote sensing datasets is required because site selection and facility layout within a base camp become difficult problems. Base camp location and layout are critical army issues due to increased environmental awareness, new construction standards to address force protection and soldier morale, and life-cycle cost. Furthermore, using multicriteria decision-making is difficult, 14 Journal of Sensors and the analysis is also complex. Training a machine with different datasets and formats is a new innovation in the field of military to modernize strategic locations for military operations. Finally, we recommend that agent-based modeling and the random forest method be used to describe the complex and intelligent decision-making behaviors of military sites, which is critical when researching military strategic site selection.

Data Availability
No data were used to support this study.

Conflicts of Interest
The authors declare that there are no conflicts of interest.