2.1 Municipal Police System

Within police administration, the municipal police system has been the subject of discourse as a mechanism to uphold the political impartiality of law enforcement since the inception of the Government Organization Act. Municipal police, in this context, have multifaceted responsibilities encompassing public safety, traffic management, security, and investigative affairs, with a particular focus on locales intricately interwoven with the daily lives of local residents. Personnel management within the precincts of police administration presents two primary quandaries. First, there is the formidable question of determining the requisite number of police personnel in consonance with the exigencies of contemporary security demands. This issue calls for a normative consensus regarding the contours of an ideal security scenario and poses intricate challenges because it relies on political resource allocation processes operating within the constraints of finite fiscal resources ^[10].

The second conundrum revolves around the judicious apportionment of the existing police force. Concerning the allocation of the available workforce, there is an expectation that the rationalization of personnel management can be enhanced, given the relatively constrained political processes in augmenting human resources ^[1]. With the advent of the Municipal Police Committee in Region G and the fully fledged implementation of the autonomous police system, the requirement of instituting a commissioned autonomous police workforce tailored to the idiosyncrasies of Region G has come into sharp focus ^[2]. The southern precinct of Area G is witnessing the proliferation of new urban developments, an influx of residents, and, notably, a surging proportion of foreign residents, accounting for 84%. In addition, the southern precinct now hosts certain military facilities alongside 35% of the residents who have relocated from specific regions. Hence, an exhaustive assessment of management imperatives is warranted.

In particular, the gravity of this study is underscored by the conspicuous shortfall in police human resources vis-à-vis the imperative of maintaining public order within Area G ^[2]. The southern region of area G boasts 559 residents per National Police Agency officer. The northern region records 539 individuals per officer, ranking them first and second, respectively (as per the 2020 Police Statistical Yearbook, as shown in Figs. 1 and 2). Fig. 1 shows the number of residents in charge per police officer, and Fig. 2 shows the number of police officers according to the city and province nationwide.

Fig. 1. Number of residents in charge per police officer.

Fig. 2. Number of police officers according to the city and province in the country.

2.2 Decision Tree

Decision trees are used widely in general data mining and decision analysis. This method finds extensive utility in artificial intelligence, machine learning, statistical analysis, and decision tree algorithms. The term 'decision tree' is commonly used to denote this method. It is used for classification or prediction purposes when dealing with categorical target variables or scenarios where case-type target values are unsuitable for the decision tree algorithm. When applying numeric variable data to the target variable, it can be used as a numeric or categorical variable. The results of data analysis through the decision tree are visually represented as a tree structure, making it highly comprehensible to analysts, which is a significant advantage. From a technical perspective, decision trees often provide classification accuracy comparable to neural network approaches, logistic regression analysis, and other classification methods while delivering more interpretable and easily explainable results.

On the other hand, decision tree algorithms can encounter challenges when dealing with data exhibiting non-discrete characteristics in the vertical/horizontal ratio of a particular variable. Unlike neural network algorithms, which consider multiple variables simultaneously, decision trees can take two main approaches: the Hill Climbing Greedy approach and the general Greedy algorithm. As with other greedy algorithms, decision trees do not guarantee optimal solutions. In addition, variations in the number of records can result in significantly different tree structures ^[8,9]

2.3 C4.5 Algorithm and CART

Supplementary devices for the blind require voice guidance or Braille output technology. Among them, voice synthesis technology produces voice signals using the frequency characteristics.

The Weka data mining tool known as J48 is based on the C4.5 algorithm. The C4.5 algorithm, proposed by Quinlan in 1993, shares similarities with the ID3 algorithm. The ID3 algorithm is a widely used tree-based classification algorithm with some limitations. The C4.5 algorithm was developed to address these limitations, including handling continuous attributes, filtering out irrelevant attributes, managing the tree depth, handling missing attribute values, and improving time efficiency in implementation.

The CART (Classification and Regression Trees) analysis technique is intricately intertwined with data mining methodologies, particularly Decision Tree Analysis. Within CART, homogeneity is pivotal, which is assessed through the Impurity function to determine homogeneous groups. Homogeneity, in this context, indicates the degree of similarity in data characteristics and types, suggesting a lack of dispersion akin to pure derivatives. The CART algorithm, a non-parametric technique, constructs a classification or regression tree based on the numerical classification of the dependent variable. This algorithm develops the decision tree guided by a split rule, where the fundamental principle is to yield the net outcome of all conceivable splits for selecting the optimal child node partition. The classification decision tree training, utilizing the CART algorithm, can be defined using a multistep process. In CART, each partition hinges on an extensive array of input variable values, often numbering in the millions. Therefore, an exhaustive exploration of all potential partitions is conducted for a specific partition ^[8].

3.1 Learning Model Design

Within the ambit of rigorous big data analytics, it is essential to curate meticulously the pivotal variables intrinsically linked to the extant workforce. These salient variables are extracted judiciously, employing an exhaustive correlation analysis to fathom their innate interplay with the prevailing count of autonomous police personnel ^[3]. The elucidation of an optimal count of security personnel, grounded in a cadre of independent variables, necessitates the development of a meticulous regression model. This model intricately weaves population size, crime incidence rates, and incident reports as robust independent variables, with the extant tally of security personnel in Seoul as the dependent variable. Furthermore, the quest for a comprehensive learning model entails the painstaking construction of a linear regression model. This model revolves around autonomous security personnel within the public security agency, assuming the role of the dependent variable while all other variables maintain their independence ^[5,6].

In the precincts of police administration personnel management, two paramount conundrums loom large. The primary quandary centers on determining an optimal police force size considering the prevailing security exigencies. The resolution of this intricate issue necessitates a normative consensus regarding the ideal security milieu and astute maneuvering within the political landscape governing resource allocation, all within the constraints of finite financial resources.

The secondary predicament hinges on the judicious allocation of the existing police workforce. In resource allocation, where human resources constitute the currency, the rationality underpinning personnel management undergoes appreciable enhancement ^[7]. This stems from the streamlined political process entailed in augmenting the workforce. The overarching objective underscores the discernment of optimal jurisdictions for each police station, facilitated by big data analytics. This discernment guides the precision of human resource allocation, ensuring seamless alignment with localized security requisites. Table 1 lists the big data analysis target and scope.

Conducting a comparative analysis of the workload per police officer based on socio-economic and demographic data, stratified by geographic regions, to identify organizations necessitating supplementary workforce. The research focuses on the areas within the purview of the Municipal Police Committee.

3.2 Data Analysis Subject and Analytic Procedure

Data analysis in this study encompasses a range of variables for evaluation. These variables include the following: the count of autonomous police officers categorized by their respective functions, such as female youth, traffic, and life safety; the tally of 112 emergency reports related to female youth, traffic, and life safety incidents; reported cases involving women; traffic accident reports; incidents related to order maintenance; lost item reports; dispatch cases; major crime incidents including rape, forced indecent assault, murder, robbery, theft, and violence; resident population statistics; female population figures; the count of single female households; data on the foreign population; statistics on the floating population; vacant housing data; information on lodging; and land use details, including public land, industrial land, agricultural land, commercial and business zones, forested areas, prior residential land, mixed-use land, and specialized zones. The data analysis procedure adhered to a structured sequence, as shown in Fig. 3:

Step 1. Data Exploration: This initial phase involved examining the status of the data and identifying significant independent variables.

Step 2. Calculation of Additional Autonomous Police Officers: In this stage, the required number of supplementary autonomous police officers for each police station under the jurisdiction of the Southern Police Agency in Area G was calculated to ascertain the appropriate staffing levels.

Step 3. Personnel Allocation: The count of additional autonomous police officers determined in the previous step (Step 2) was allocated by their specific functions, which include traffic, life safety, and female youth-related duties.

Fig. 3. Task analysis process.

3.3 Decision Tree Modeling and Prediction Method

A decision tree model was meticulously constructed, using the number of municipal police officers as the dependent variable while treating all other variables as independent factors. Corresponding to the regression model, this decision tree model processes regional data by inputting it into the previously trained model. One of the pivotal advantages of the decision tree model is its superior explanatory capacity, which examines why a specific number of police officers is deemed necessary. This model adeptly predicts the suitable staffing levels across different functions while utilizing the dependent variable of the learned.

The model formulation, alongside the analysis values, was defined and examined. These results were then generated and subjected to analysis using Python code. The model considers the following variables. In addition, the modified R-squared value, denoted as R$^{2}$ = 0.9210, was considered a variable factor.

Analysis value = 0.000414 * number of reported cases - 0.50987 * robbery (cases) + 0.000866 * youth population (individuals) - 0.000273 * elderly population (individuals) + 0.017718 * violence (cases) - 0.000224 * child population (individuals) + 0.000748 * Female population (individuals) - 0.000323 * Total population (individuals) - 0.026015 * Theft (cases) + 0.829467 * Murder (cases) + 34.626932.

Preprocessing process: The collected data underwent a series of preprocessing steps using various tools and techniques. Fig. 4 presents the preprocessing procedure. Initially, the original data in formats, such as csv or xls, were imported into the chosen tool. (Data Loading) The second step involved a meticulous assessment of the uniqueness. By examining data uniqueness based on pre-existing and new data KEY values, this stage determined if the Records value of Results-Unique-Out-Duplicates equaled 0. If not, it indicated that existing or new data warranted further examination. (Uniqueness Review) The third step evaluated the data redundancy, focusing on the existing/new data KEY value. If the Records value of Results-Join-Out-Join was not zero, it necessitated a thorough inspection of duplicated data within the existing/new data. (Data Redundancy Review) The fourth phase facilitated data verification and collection through data integration (Join). This step gathered existing and new data, not duplicated, by leveraging the KEY values. (Data Integration-Join) Finally, the refined and streamlined data was securely saved, completing the preprocessing pipeline. (Data Preservation)

Fig. 4. Serial communications code in TensorFlow.

3.4 Model Learning with Optimal Branching Point

The dataset from the S Police Agency was employed for model learning, while the data from the Southern Police Agency in the G region was used for practical application. This model was developed to increase the number of police officers to a level akin to that of the S Police Agency. This was done assuming an appropriate number of municipal officers were deployed in the S Police Agency. The personnel distribution was grounded in decision tree modeling principles. A multi-label modeling approach was adopted in the learning phase, encompassing three dependent variables: traffic, life safety, and the current female youth ratio. A decision tree model was selected to capture the relationships between the independent and dependent variables, expressed as a combination of discernible rules.

Consider the relationship between the key variables and the number of Municipal police. The connection between the key variables and the number of autonomous police was represented visually using scatter plots to facilitate data analysis. In addition, the Pearson correlation coefficient was computed to quantify this relationship. Pearson's correlation coefficient indicates the linear association between two variables, with values ranging between ${-}$1 and 1.

4.1 Analysis Model

With the full-scale implementation of the municipal police system, the imperative to establish an operation for municipal police personnel aligned with regional characteristics has come to the forefront. An analysis of the relationship between the defined variables and the total number of Municipal police officers revealed a consistently significant positive correlation. This is particularly pertinent in areas characterized by factors such as rapid new town development, population influx, a high proportion of foreign residents based on the latest statistics from KOSIS (as of March 2022) at 84%, the relocation of U.S. Forces Korea, a significant presence of North Korean defectors supported by settlement policies from the Humanitarian Cooperation Bureau of the Ministry of Unification (as of June 2022) at 35%, and others.

These trends underscore the escalating demand for human resource management strategies that consider the specific requirements of local security. In particular, the existing shortage of police personnel in proportion to the increasing need for local security emphasizes the urgency of implementing efficient and regional-tailored workforce management practices. The demand for human resources management reflects the demand for local security, which has increased rapidly. In particular, there is a shortage of police staffing to meet the demand for local security. Hence, efficient human resources management tailored to the region is urgent.

The Municipal Police Cooperation Division seeks recommendations for an efficient workforce allocation strategy that meticulously reflects the regional characteristics within the framework of municipal police personnel management. This necessitates significant structural adjustments, including the recalibration of task composition ratios. To this end, the aim was to advance the field of human resources planning by transitioning from intuitive personnel management practices to an objective and scientific approach, achieved through rigorous data analysis. The analysis will leverage data provided by police agencies and local governments to examine the regional attributes and workload distribution at the administrative ``dong'' level and perform a comprehensive comparative analysis of the current human resource management status across individual police stations against the requisites for efficient workforce allocation.

4.2 Regression Model and Results

A precise prediction was derived by constructing a linear regression model using the number of Municipal police officers in the S-area Metropolitan Police Agency as the dependent variable, with all other pertinent variables serving as independent factors. This learned model was leveraged to predict the required number of Municipal police officers for each region. The exact staffing deficits were determined by calculating the difference between these predictions and actual headcounts. The analysis showed that, except for certain areas, an average increase of 43.27% in the municipal police force is indispensable for meeting the heightened security demands. On the other hand, when considering all independent variables in the analysis, there is a significant chance that a surplus of police officers might be necessary due to potential redundancies or overlapping factors among these variables. Fig. 5 presents the results of this correlation analysis.

Fig. 5. Correlation analysis results.

4.3 Experiment Crystal Tree Model and Analysis

A decision tree model was trained using the number of municipal police officers as the dependent variable and all other variables as independent variables. Similar to the regression model, the required number of personnel was calculated by inputting regional data into this trained model. The decision tree model, known for its superior explanatory power, facilitates a clear understanding of why additional police officers are required. Interpretation Method: The interpretation involved comparing the average number of municipal police officers satisfying specific criteria. For example, in 2020, the average number of municipal police officers within the Seoul Metropolitan Police Agency satisfying the following criteria was calculated: no more than 24,099 reported cases, no more than 73 cases of rape, no more than 798 cases of violence, and no more than 1,020 cases of theft. On the other hand, when applying the same conditions to the GW Police Station and UI Police Station, they had only 52 and 53 officers, respectively. Hence, they need to increase their personnel by 35 and 34 officers, respectively. In 2020, when the number of reported cases exceeded 24,099, but the number of rape or forced molestation, violence, and theft cases remained at 73 or fewer, 798 or fewer, and 1,020 or fewer, respectively, the average number of Municipal police officers in S region was 76.75. In 2021, the average number of municipal police officers in Seoul was 108 when the number of reported cases exceeded 61,051, the number of theft cases was less than 880, the number of violence cases exceeded 798.

The appropriate number of personnel by function was predicted based on the predicted values from the learning model, considering the learned patterns. Fig. 6 shows the distribution of the modified R-squared values based on the number of independent variables used. The optimal performance was achieved when utilizing 10 independent variables.

Table 2 lists the actual number of municipal police officers and the appropriate number of police officers predicted using the above model, which aligns with the number of municipal police officers.

This particular model omitted the calculation of the optimal personnel count via machine learning and relied solely on data from the Southern Police Agency in the G region. While several variables were consistent with those applied in the main body of the study, the variables identified as significant differed because of the variations in the dataset. Table 3 lists the allocation of the suitable number of police officers according to function using the acquired decision tree model.

The analysis results utilizing the data from all police stations are as follows: All variables, except for robbery cases, exhibited a significant positive correlation with the count of municipal police at the 0.05 significance level. In particular, the count of reported cases, incidents of theft, cases of rape and forced molestation, and the female population showed a robust positive correlation with the number of municipal police officers. Table 4 lists the results of correlation analysis between the major variables and municipal officers when applying all police station data. Fig. 7 shows the correlation between the major variables and the number of autonomous police in the G.S region.

Fig. 6. Modified R-squared distribution according to the number of independent variables.

Fig. 7. Correlation between the major variables and the number of municipal police. (G.S region)

Table 2. Experimental results with the object recognition system.

Table 3. Appropriate number of police officers according to the police station.

Table 4. Results of correlation analysis between the major variables and municipal officers: All police station data applied.