MAINTENANCE PREDICTION USING MACHINE LEARNING TECHNIQUES

Abstract

Maintenance of software is a repetitive stage in the software development life cycle. It starts just after the software product is deployed to the customer and this phase ends when the product is no longer in use or its been outdated. There are different exercises carried out in software maintenance phase, for example, the addition of advance feature, deletion of unwanted feature, error correction, adaption to new condition and so forth. Maintainability of software is the quality property of the product software which decides the path with which these adjustments or modifications can be performed to give better outcome. The principle target of this report is to analyse machine learning technique, its use along with genetic algorithm particularly with Ward neural network or NNEP GRNN(Neural Network Evolutionary programming) for maintainability prediction of software and tried to improve the performance and efficiency measures obtained from previous studies using different machine learning techniques but with improvised approach using object oriented metrics only, Here In this work the datasets which are used is same as other study because it is more accurate and useful for software maintenance and hence we follow the same datasets and obtain the result with better efficiency. This procedure is connected to appraise practicality on two diverse case studies, that is Quality Evaluation System(QUES) and User Interface System (UIMS). In this approach, for updating the weight amid learning phase, genetic algorithm is utilized. Later, We plot graph and table and depict its performance with various other machine leaning techniques such as General Regression Neural Network, Decision table/tree etc. Neural network along with evolutionary/genetic programming is analysed as the best model for prediction than comparing with other ML techniques like ward neural network (i.e neural network with different slabs in the hidden layer). Neural network along with genetic algorithm i.e the hybrid approach that we used with PCA is capable in reducing MSE (Mean squared error) and MMRE (maximum magnitude of relative error) into a less error value so that this approach can be used further for new data or software industry as well.