Integrating digital forensics and machine learning for fraud detection in Nigerian deposit money banks: a PLS-SEM and artificial neural network approach

A PLS-SEM and Artificial Neural Network Approach

Abstract

Purpose: This study examines the association between digital forensic capabilities and fraud detection effectiveness in Nigeria’s five largest deposit money banks—United Bank for Africa Plc, Zenith Bank Plc, Access Bank Plc, First Bank Nigeria Limited, and Guaranty Trust Bank Plc. Specifically, it evaluates the relative contributions of network forensics, cloud forensics, mobile forensics, and Internet of Things (IoT) forensics to fraud detection effectiveness.

Methodology: A cross-sectional survey design was adopted, with data collected through structured questionnaires administered to staff of the selected banks. Of the 306 questionnaires distributed, 243 valid responses were obtained, representing a 79% response rate. Data were analyzed using Partial Least Squares Structural Equation Modeling (PLS-SEM) to test hypothesized relationships and Artificial Neural Network (ANN) analysis to assess predictive importance. PLS-SEM analysis was conducted using JASP version 0.95.3, while ANN estimation employed a multilayer perceptron model in SPSS version 27.

Results and conclusion: The PLS-SEM results indicate that digital forensic capabilities jointly explain a substantial proportion of variance in fraud detection effectiveness with all R² > 0.75. Cloud forensics (β = 0.515) and network forensics (β = 0.500) exhibit the strongest associations, followed by IoT forensics (β = 0.456) and mobile forensics (β = 0.449). ANN results corroborate these findings, ranking cloud forensics as the most important predictor (normalized importance = 100%), closely followed by network forensics (99.7%), mobile forensics (85.3%), and IoT forensics (79.8%).

Implication of findings: The study implies that bank management should prioritize investment in cloud and network forensic infrastructures, complemented by targeted training and process integration, to enhance fraud detection performance. However, the cross-sectional and self-reported nature of the data limits causal inference, indicating the need for future longitudinal studies using objective fraud performance indicators.