Quantitative Analysis and Design For Software Rework Reduction Using Genetic and Deep Learning Techniques

Abstract

Historical data collection provides knowledge for technical analysis and predicts the fault in the early stage. Maintaining historical data helps to improve development and reduce the rework. Finding new patterns from data forms the structure of algorithms and assists in prediction and feature extraction. Data extraction plays a vital role in all prediction and classification related problems. Data extraction will be successful only by finding the root cause of a fault, and the genetic algorithm makes this process optimal. Fault taxonomy with genetic nature will help to handle similar cases and suggest future actions. This classified fault domain knowledge learns the patterns and identifies the effective strategies for hybridization. Fault prediction is the process of detecting a fault in the software life cycle phases. Various prediction and classification methods establish and evaluate software fault prediction. These approaches provide relatively promising prediction results for different software projects across the industry. Some hybrid model improves the prediction performance of software faults. However, existing study reports hypothesize the efficiency of algorithms might vary on applying different performance measures. This research work improves the performance using hybrid classifiers (K-means clustering, SVM, and Genetics) and ensembles (deep learning and data science) in software fault prediction using multi-criteria decision making. This work proposes a novel methodology for predicting, classifying, and evaluating software defect modules. ii This work consists of fault feature analysis and filtering by hypothesis testing. Also, it includes the feature diagnosis and reduction by genetic programming. Finally, the GKS technique measures the feature performance. The automation of this research work includes defect predictor, defect taxonomy, and rework evaluator using genetic deep learning and data science techniques. An experimental study designed using open-source and real-time repositories help in th