Harnessing Machine Learning Insights for Unveiling the Complexity of Corruption Behavioral Dynamics

The ingenious artificial intelligence (AI) tactics are fighting against the menace of corruption prevailing in society, creating transparency by reducing asymmetric information, automating administrative regulations, and detecting unethical, scam, and fraudulent activities. Breaking away from conventional approaches, the present research incorporated the time series prediction strength of machine learning via nonlinear autoregressive exogenous (ARX) neuro-architecture trained with the Levenberg Marquardt optimization (LMO) algorithm, i.e., nonlinear ARX-LMO, for solving the mathematical model of corruption dynamics within a population (CDP) comprised of the five state variables regarded as susceptible to corruption, exposed to corruption, individuals performing corruption, recovered from corruption, and the honest individuals. The synthetic data are produced utilizing the computationally efficient and reliable Adams numerical approach to stimulate the realistic population dynamics within the generalized CDP model. A comprehensive set of case studies along with the diverse scenarios formulated with the variations of key parameters in terms of the proportion of susceptible joins the honest group, exposed individual’s conversion rate to a corrupted group, probability of corruption risk per interaction, recovered individual’s conversion rate from a corrupted group, and the proportion of exposed individuals joining the corrupted group, are formulated to capture the complexities concerning real-world corruption dynamics. The evaluation based on the network performance and comparative solutions results confirms the reliability of the proposed nonlinear ARX-LMO scheme in terms of robustness, precision, and accuracy, depicted via error distribution, serial correlation analysis, and scrutiny of the mean square error (MSE) metric.