Adaptive Decision-Support System Model for Automated Analysis and Classification of Crime Reports for E-Government

This study explores the potential of text analysis and classification techniques to improve the operational efficiency and effectiveness of e-government, particularly within law enforcement agencies. It aims to automate the analysis of textual crime reports and deliver timely decision support to policymakers. Given the increasing volume of anonymous and digitized crime reports, conventional crime analysts encounter challenges in efficiently processing these reports, which often lack the filtering or guidance found in detective-led interviews, resulting in a surplus of irrelevant information. Our research involves the development of a Decision Support System (DSS) that integrates Natural Language Processing (NLP) methods, similarity metrics, and machine learning, specifically the Naïve Bayes' classifier, to facilitate crime analysis and categorize reports as pertaining to the same or different crimes. We present a crucial algorithm within the DSS and its evaluation through two studies featuring both small and large datasets, comparing our system's performance with that of a human expert. In the first study, which encompasses ten sets of crime reports covering 2 to 5 crimes each, the binary logistic regression yielded the highest algorithm accuracy at 89%, with the Naive Bayes' classifier trailing slightly at 87%. Notably, the human expert achieved superior performance at 96% when provided with sufficient time. In the second study, featuring two datasets comprising 40 and 60 crime reports discussing 16 distinct crime types for each dataset, our system exhibited the highest classification accuracy at 94.82%, surpassing the crime analyst's accuracy of 93.74%. These findings underscore the potential of our system to augment human analysts' capabilities and enhance the efficiency of law enforcement agencies in the processing and categorization of crime reports.