Hybrid REMEDIAL-Resampling Strategy for Audit Findings of Financial Statements Data on Multi-Label Imbalance Scenario

The Audit Board of The Republic of Indonesia (BPK RI) is the institution responsible for auditing accountability of state finances. However, limited human resources, a large number of entities, and limited audit time remain obstacles to the implementation of audits, which makes accurate sample selection important for audit planning. Previous studies have applied multi-label classification approaches to determine audit samples by predicting audit opinions and findings. The multilabel classification that has been carried out still has an imbalance of labels, which was addressed by removing minority labels that may contain important information. This study proposes the application of the REMEDIAL resampling method combined with other techniques such as MLROS, MLeNN, and MLSMOTE to overcome label imbalance without removing minority labels. A multi-label classification model was built using a problem transformation approach with the XGBoost base classifier. The results showed that the combination of REMEDIAL and MLSMOTE was able to improve label distribution while also improving model performance. The application of this method presents robust model performance with an accuracy value of 0.73, precision of 0.75, recall of 0.94 and an F1 score of 0.83. With the achieved performance, this model is expected to be used by BPK RI auditors to predict findings from local government financial statement audits in region X, thereby assisting in more accurate audit planning.