A double-stacked ensemble classification approach for crime against women in India

The rising frequency of crimes against women necessitates the development of the accurate and effective methodology to enable targeted interventions and preventive measures. This research utilizes a dataset prepared from the National Crimes Record Bureau’s 2022 report on crimes against women. In this study, we investigated the application of advanced ensemble classification techniques to analyze the crime against women dataset. The spatial distribution of aggregated crimes is illustrated using a map of India, providing a visual representation of how crimes against women are distributed across the country. Our proposed approach employs a hybrid ensemble model that integrates the base learners as Light Gradient Boosting Machine, CatBoost, and AdaBoost, applied with the novel hybrid meta-learner. The novel hybrid meta-learner incorporates Random Forest and CatBoost as base learners, with logistic regression serving as meta-learner. By stacking these base learners and applying the novel hybrid meta-learner, this approach capitalizes on the strengths of each technique, overcoming their individual limitations and enhancing overall classification performance. We evaluate performance of baseline models, ensemble models, and our proposed hybrid ensemble model using metrics, accuracy, precision, recall, and F1-score. Our comparative analysis showed that the proposed hybrid model surpasses both baseline and conventional ensemble methods in classification accuracy and reliability. Additionally, the efficiency of the Meta_RCL model is validated through cross-validation, achieving an accuracy of 0.969, precision of 0.969, recall of 0.968, and F1-score of 0.968 which are the highest compared to other models. This model’s versatility allows it to be applied to other crime datasets, aiding law enforcement agencies, policymakers, and social organizations in identifying patterns, trends, and risk factors.