Optimizing fault prediction in software based on MnasNet/LSTM optimized by an improved lotus flower algorithm

Software quality and reliability are very important problems in the field of software production. Software error and defect detection technology is one of the most important research goals in the field of software system reliability that prevents software failure. Therefore, the performance of the defect prediction model in order to accurately predict defects is important in improving and effectiveness of models. In this paper, an attempt has been made to present a hybrid and efficient classification model based on deep learning and metaheuristic models for predicting defects of software. The basis of the suggested model is utilizing a combination of MnasNet (for extracting the semantics of AST tokens) and LSTM (for keeping the key features). It has been improved with the help of an improved variant of Lotus Flower Algorithm (ILFA) so that appropriate coefficients and acceptable results can be produced with the optimization power of metaheuristic algorithms and the learning power of the network. For evaluating the results of the suggested model, the model is applied to a practical dataset and the results are compared with some different methods. The new combined model worked best for the Xerces project, reaching 93% accuracy, which was much better than other models. It also performed well on different projects, improving accuracy by 3.3% to 7.9% after cleaning the data and fixing the issue of uneven class sizes. The results indicate that the proposed model can achieve the highest values ​​of efficiency.