A Hybrid Machine Learning and Regression Approach for Validating a Multi-Dimensional Crime Index in the Context of Crime Against Women

Violence targeting women has endured since ancient times, encompassing a spectrum of offenses ranging from psychological anguish to physical and sexual assault. This study introduces a crime index rooted in diverse categories that directly or indirectly contribute to fostering criminal intentions. A composite weighted index, comprising four sub-indexes focusing on Health, Socioeconomic status, Education, and Judiciary, was created. The stability and homogeneity of the index were assessed using reliability testing. Validation of the proposed index was carried out through comparative analysis of baseline and ensembled models. The hybrid model was proposed by combining multiple linear regression and robust regression techniques with random forest and stochastic gradient descent as meta-regressors. The models were assessed by the evaluation metrics MAE, RMSE, and MAPE. The findings indicate that the Index demonstrates strong reliability, supported by a significantly high correlation. The ensemble hybrid model approach, effectively captures the variance of the model, with less error when compared with the baseline models but the homogeneous ensemble approach proved to be the best with minimum error. Women in less developed regions with extreme geographical conditions are at a higher risk of falling prey to victimization. The Index and its statistics conclude that social factors significantly contribute to the occurrence of violence against women. These findings hold the potential for informing enhanced strategies to curtail the menace.