Integration of IndoBERT as a Feature Extractor with Machine Learning and Deep Learning Algorithms for Quality Management System Audit Findings Classification

The SNI ISO 9001:2015 audit process faces significant challenges in accurately classifying non-conformity findings due to the standard's complexity. Misclassification leads to ineffective corrective actions and recurring quality issues. This study aims to develop an Artificial Intelligence-based text classification model to automate the mapping of Indonesian-language audit findings to their respective clauses, leveraging IndoBERT's linguistic capabilities. This research adopts a quantitative approach by integrating the IndoBERT Pre-trained Language Model as a feature extractor with two modelling approaches, the traditional Machine learning algorithms such as Support Vector Machine (SVM), XGBoost, Random Forest and the Long Short-Term Memory (LSTM) Deep learning architecture. IndoBERT generates contextual semantic representations from the finding texts, which are then used as input features for two modelling approaches, traditional Machine learning algorithms such as SVM, XGBoost, and Random Forest and the Long Short-Term Memory (LSTM) Deep learning architecture, aimed at capturing sequential dependencies within the text. Model performance was evaluated and compared against conventional (SVM) and pure Deep learning (LSTM) baselines. The experimental results definitively show that the IndoBERT integration strategy is significantly superior. The IndoBERT - LSTM model was established as the absolute best model, achieving the highest Accuracy of 0.90 and an F1-Score of 0.90. This performance represents an improvement of 45.16% over the pure LSTM baseline and 26.76% over the SVM baseline. Overall, the IndoBERT - LSTM model provides the most accurate and consistent solution for automating the classification of audit findings.