<![CDATA[Purpose – This study develops and preliminarily evaluates an interpretable hybrid anomaly-monitoring prototype for Accounting Information Systems (AIS) in resource-constrained financial institutions. Internal-control effectiveness is operationalized narrowly as the prototype’s ability to identify and prioritize anomalous transactions for timely human review, rather than as organization-wide control effectiveness. Method – A simulation-based prototype experiment uses a labeled synthetic dataset of 852 transactions. Three univariate detectors applied to transaction value Z-Score (|Z| > 2), Interquartile Range (1.5×IQR), and Isolation Forest are integrated through an explicit rule-based aggregation. Performance is assessed against synthetic ground-truth labels using a confusion matrix and imbalance-aware measures. Findings – The dataset contains 809 normal and 43 anomalous observations. The hybrid classification produced TN=785, FP=24, FN=5, and TP=38, yielding accuracy=0.966, precision=0.613, recall=0.884, F1-score=0.724, specificity=0.970, and balanced accuracy=0.927. The results demonstrate preliminary classification feasibility while indicating a need to reduce false positives. Research implications – The prototype provides interpretable anomaly signals to help authorized reviewers prioritize transactions for further examination; it does not autonomously determine fraud. Originality– The study integrates transparent statistical and machine-learning signals into an AIS-oriented early-warning architecture, defining internal-control effectiveness specifically at the transaction-monitoring level within resource-constrained financial institution monitoring environments.]]>
