A Novel Adam Hypertuned Artificial Neural Network Using Autoencoder Network Approach to Improve Software Defect Prediction Accuracy

Defect prediction is a very active area in the software engineering field. It is crucial to bridge the gap between software engineering and data mining to ensure the programme's success. Predicting software flaws helps find faults in the code before testing is done. Cluster analysis, statistical approaches, mixed algorithms, neural network-based metrics, black box testing, white box testing, and machine learning are only some of the methods used to investigate the software effect area while trying to forecast defects in software. In order to improve the accuracy of deep learning classifiers for defects forecasting, this study makes a novel contribution by using feature selection for the first time. This research was conducted with the hope of enhancing the accuracy with which errors may be predicted in five NASA data sets: CM1, JM1, KC2, KC1, and PC1.Here initially the data was retrieved and processed using rounded mean regressor interpolation approach. Then for selecting feature information grain methodology was used. Dimensionality Component Analysis (DCA), Self-Regulating Component Analysis (SRCA), and Non-Negative Linear Matrix Factorization (NNLMF) were used to extract features from the recovered data . In order to improve upon previous techniques of defect prediction, we combine the factorization selection approach with the deep learning-based adam hypertuned ANN using autoencoder method. All of the tests were run in a python environment. This research shows that, in comparison to the currently used mechanisms, defect prediction accuracy may be increased by the application of feature selection.