Abstract
An organizational audit plays a critical role in identifying risk factors, evaluating resource compliance, and supporting informed managerial decision-making. However, traditional audit approaches are increasingly limited by their manual, time-consuming nature and their inability to process complex, multidimensional organizational data efficiently. To address these limitations, this study proposes an intelligent audit risk assessment framework based on machine learning (ML) and deep learning (DL) models. The framework is evaluated on a real-world dataset comprising 773 organizational units. A multi-stage methodology is applied, including data preprocessing, normalization, feature selection, class balancing, and model development. A range of ML and DL models, including Naive Bayes, support vector machine, decision tree, random forest (RF), XGBoost, dense neural network (DNN), convolutional neural network (CNN), long short-term memory (LSTM), recurrent and hybrid CNN-LSTM, and LSTM-gated recurrent unit (LSTM-GRU) models, are implemented. The performance is evaluated using accuracy, precision, recall, and F1-score. Experimental results show that ensemble models, particularly RF and XGBoost, achieve stable, well-generalized performance with test accuracies up to 0.957 and F1-scores up to 0.958, while a DNN and recurrent neural networks demonstrate competitive performance. Overall, the proposed framework demonstrates the strong practical potential of ML and DL models for an organizational audit.