Hidden Markov Model with Baum–Welch Algorithm for Fraud Detection in Academic Business Processes

Abstract

This study investigates fraud detection in academic business processes, where fraudulent behavior is often latent and embedded in sequential activities recorded in event logs. Fraud is defined as deviations from established academic procedures, including course registration (KRS) without payment of the Operational Education Fee (DOP), role violations, and irregular activity sequences. To address this problem, this study employs a Hidden Markov Model (HMM) to model the sequential and hidden nature of academic process behavior. The dataset consists of academic event logs from the odd semester of the 2024/2025 academic year, which were processed through data preprocessing, event-log transformation, feature construction, and rule-based labeling. Violation-based features, including wdecision, skippedactivity, wresource, wduty, wthroughput, and wpattern, were used as observation symbols in the HMM. Model parameters were estimated using the standard Baum–Welch algorithm, while the Viterbi algorithm was applied to infer the most probable hidden state sequence for each case. Model performance was evaluated using a confusion matrix and standard classification metrics. Experimental results show that the proposed approach achieves an accuracy of 96%, a precision of 87%, a recall of 100%, and an F1-score of 93%. These results demonstrate that HMM is effective for detecting fraudulent patterns in academic business processes by capturing sequential deviations that are difficult to identify using non-sequential models.