An Analytical Study of ML Models for Behavioral Analytics in Catching Suspicious Payroll Activity

Machine learning methods are essential for enhancing behavioural analytics, enabling automatic pattern discovery and adaptive learning.Supervised learning models benefit from detecting known instances of fraud when labeled data are available.In contrast, unsupervised and semi-supervised methods can detect previously unknown anomalies without clearly defined labels indicating whether the data are fraudulent.Furthermore, deep learning models, including autoencoders and recurrent neural networks, can learn complex, nonlinear, and temporal dependencies in payroll data.All these features enable machine learning-based systems to keep up with changes in payroll practices, the workforce, and fraud strategies, thereby enhancing detection accuracy and resilience.In practice, data imbalance, feature selection, model interpretability, scalability, and concept drift have a significant impact [7].Various machine learning models have different strengths and weaknesses based on the features of payroll data and operational limitations [8] [9].Consequently, an analytical study should be conducted systematically to determine the effectiveness of ML models in behavioural analytics for detecting suspicious payroll activity. A. Structure of the PaperThis paper is organized as follows: Section II payroll activity and suspicious activity analysis.Section III Behavioral analytic framework for payroll monitoring Section IV.Machine learning model for behavioral analysis.The literature review is presented in Section V, and the conclusions and prospective research are discussed in Section VI.