Providing a Framework for Assessing and Evaluating Network Data Studies in the Fight Against Social Anomalies

purpose: In today's world, every society is struggling with the challenges of crime and its control. Committing a crime represents one of the most significant social harms, necessitating the involvement of police and judicial institutions for effective management. Assessing, evaluating, and identifying crime patterns, as well as detecting and preventing criminal activities, have been the focal points of judicial and law enforcement agencies since ancient times. Consequently, crime prevention is consistently prioritized over crime detection after it occurs. Various methods have been introduced for crime detection, including a wide range of innovative computer techniques. Data mining is regarded as one of the most effective tools for data and information analysis in the field of crime detection. Many effective parameters are available for analyzing crime through data mining methods. Data mining serves as a valuable tool for examining crime data warehouses, helping to extract hidden knowledge within them. The application of data mining techniques, along with various machine learning methods, can yield significant benefits in identifying, predicting, and preventing crime in any society. Diagnosing, predicting, and preventing crime through data mining represents a progressive approach supported by statistical methods, psychology, artificial intelligence, criminology, machine learning, and database technologies. Consequently, the primary objective of this research is to establish a framework for analyzing digital network data in the battle against social anomalies, particularly the crime of theft. In this research, materials related to the crime of theft have been categorized into three areas: crime identification, prediction, and prevention. Additionally, the application of data mining methods has been explored within these domains.
Methodology: This research is applied in terms of its purpose and was conducted using documentary methods, content analysis, and data mining. The statistical population for this study consists of information related to theft crimes recorded by law enforcement and police organizations in 2019. Initially, the relevant data were collected in the form of documents and subsequently selected using the content analysis method to facilitate further analysis. Patterns of theft crimes were identified based on various factors, including the type of crime (such as home and car theft), entry location, entry method, search techniques, and residential area. This study utilized digital network data analysis tools and methods, specifically data mining, for classification and validation purposes. Additionally, clustering techniques, such as k-means, were employed to identify different forms of theft crimes. Classification algorithms, including neural networks, Bayesian rules, Bayesian navigation, and support vector machines, were used to predict theft crimes. The primary data analysis tool utilized in this research was Excel software.
Findings: The findings indicate that the season in which a theft occurs positively correlates with the month of the crime. Additionally, the method used to enter the crime scene shows a positive correlation with the method used to exit the scene. Furthermore, the accuracy of the Bayesian model in predicting and detecting the type of crime is 0.412. The model demonstrates the highest prediction accuracy for home thefts at 73%, while the lowest prediction accuracy for thefts from private locations is 27%. The results are expressed as a percentage. Additionally, the accuracy of each technique employed has been compared. The ROC findings indicate that the accuracy of the Bayesian methods and the Multilayer Perceptron (MLP) neural network, as well as the support vector machine, in predicting theft from public places is higher than for other types of theft. Conversely, in predicting home theft, these methods demonstrate lower accuracy compared to other thefts. Furthermore, the prediction accuracy of the support vector machine method, at approximately 91%, surpasses that of Bayesian methods (around 73%) and neural networks (about 90%) in predicting theft from public places. Moreover, the ROC chart for the support vector machine indicates that it is 7% less accurate in predicting pickpocketing compared to other thefts.
Conclusion: The results indicated that the Bayesian rules method is the most effective for detecting and predicting patterns of theft crimes. It demonstrates higher accuracy compared to other methods. Specifically, for predicting thefts from public places, the support vector method is recommended. As a result of the findings from data mining techniques, it is recommended that police and judicial organizations utilize a combination of data mining and artificial intelligence methods to enhance the detection and identification of thieves, particularly those with varying criminal histories related to theft. This approach aims to improve the accuracy of the information obtained about these individuals. Furthermore, based on the data mining techniques discussed in this research, an expert intelligent system can be developed to predict the likelihood of future criminal attempts by offenders.