A Performance Of Robust Hybrid Methodology: A Forensic Crime Case Study

Abstract

Background: This study aims to showcase an effective technique for variable selection using established Multiple Linear Regression (MLR) models. Furthermore, the study aims to validate these selected variables using multilayer feed-forward neural network (MLFFNN) models and enhance the analysis by incorporating contour plots and surface plots as visual tools. Initially, all chosen variables will undergo the bootstrap methodology to assess their significance and screen for relationships. Objective: The primary aim of this study is to create, standardize, and validate a hybrid model that integrates multiple linear regression, multilayer feed-forward neural networks, surface plot methodology, and contour plots. The model will be implemented using the R software, which offers a comprehensive modeling approach and various diagnostic tools. These diagnostic tools will assist researchers in accurately interpreting the results and obtaining optimized outcomes. Material and Methods: Approximately 200 simulated data points were utilized to establish the methodology for this study. Advanced computational statistical modeling techniques were employed to assess the data characteristics of various variables in this retrospective analysis. These variables encompassed aspects such as total victim count, gender, age, marital status, presence of adults and children in the household, burglary and sexual victimization, and victim reporting. The case study was devised and executed using the R-Studio program and corresponding syntax. Results: The statistical analysis revealed that regression modeling outperforms the R-squared and mean-square error tests in most scenarios. The researchers observed that when the data was divided into two sets for training and testing, the hybrid model approach exhibited significantly superior predictive capabilities for the experimental outcome. To determine the validity of the variables, the well-established bootstrap-integrated MLR approach was employed. In this case, seven characteristics were considered: gender (β1: 2.0882e+00; p < 0.05), age (β2: -5.8824e-02; p < 0.05), marital status (β3: 3.8235e-01; p < 0.05), adult in the household (β4 1.6176e+00; p < 0.05), burglary victim (β5: 2.5588e+00; p < 0.05), the sexual victim (β6: -2.3529e-01; p < 0.05), and victim's report (β7: -3.2353e-01; p < 0.05). The linear model in this scenario yielded a predicted mean square error (PMSE) of 4.261. While the predicted mean square error for the neural network is 0.19099. Conclusion: The main aim of this research is to develop and thoroughly assess a hybrid approach that combines bootstrapping, multilayer feedforward neural network, multiple linear regression, and contour plot and surface plot techniques. The methodology involves creating and presenting R syntax to ensure researchers have a comprehensive understanding of the approach. The statistical analysis conducted using R software in this study demonstrates that linear regression modeling surpasses other methods in terms of accuracy measures and the Mean Square Error value. As a result, the findings of the study strongly support the superiority of the hybrid model technique, contributing to a deeper understanding of its significant impact on the outcomes in this particular case.