A Decision Support System (DSS) for Fraud Detection Using Genetic Support Vector Machine (GSVM)

Abstract

Purpose: This study developed a Decision Support System (DSS) for fraud prediction using a Genetic Support Vector Machine (GSVM), a hybrid model that employs a Genetic Algorithm (GA) to optimize Support Vector Machine (SVM) hyperparameters (C and γ) on financial ratios extracted from MachameRatios®. Research Methodology: A quantitative ex post facto design using data from 75 purposively sampled quoted manufacturing firms across six sectors over an 11-year panel (2011–2021) was assessed. Class imbalances were handled via Synthetic Minority Oversampling Technique (SMOTE), data were labelled via a performance-based threshold, and the hybrid framework was benchmarked against Decision Tree, Bayesian Networks, and Naïve Bayes classifiers. Results: The GSVM with an Radial Basis Function (RBF) kernel achieved an optimal 10-fold cross-validation accuracy of 87.91%, 89.4% sensitivity, and an Area Under the Receiver Operating Characteristic Curve (AUC-ROC) of 0.92, significantly outperforming the standard baseline models. Conclusions: The study concludes that the wrapper-based GSVM provides a highly accurate and parsimonious nonparametric pipeline for financial screening, establishing a mathematically robust foundation for automating corporate surveillance in regional markets. Limitations: This study relied solely on the financial ratio architectures of the sampled manufacturing firms. Contributions: The system addresses data imbalance via SMOTE and demonstrates that GA-optimized feature selection outperforms other methods.