ENHANCING MOBILE APP QUALITY THROUGH DEFECTS PREDICTION USING CNN-BILSTM MODEL
Mobile applications are widely used, but undetected defects can negatively impact user experience and business reputation. Ensuring defect detection before release is crucial for maintaining software overall quality and reliability. Traditional defect prediction models struggle to handle the complexity of mobile applications. These models rely on predefined rules and handcrafted features, which often results in poor generalization and high false positive rates. Existing deep learning models also face challenges in capturing both spatial and temporal relationships in mobile app code. This study developed a hybrid CNN-BiLSTM model to improve defect prediction accuracy. The proposed approach included feature selection, data preprocessing, and model training using a dataset from COMMIT GURU. The model was evaluated using ten Android application datasets and achieved an average accuracy of 95% and an AUC of 93%, significantly outperforming previous models, which attained only 69% accuracy with the same AUC. These results validated the effectiveness of the proposed CNN-BiLSTM hybrid model in accurately identifying software defects in mobile apps and highlighted its potential for improving software reliability. Future work should focus on improving the model�s generalizability across diverse datasets and optimizing its computational efficiency to enhance scalability and real-time applicability.
Authors : Muhammad, A.A., Umar, K. and Agaie, A.I.
Category : Open Access Volume (Issue) : 11(2) Date Uploaded : 9th January 2026