Search academic texts

Information × Registration Number 0223U001980, 0121U109527 , R & D reports Title Development of information technology for software reliability assessment and forecasting by machine learning methods popup.stage_title Head Yakovyna Vitaliy S., Доктор технічних наук Registration Date 06-02-2023 Organization Lviv Polytechnic National University popup.description2 For research, we used a data set obtained by merging data sets KC1, KC2, PC1, CM1, JM1 from the PROMISE Software Engineering repository, which contained data on software module testing and 21 code metrics. The ANOVA F-value, Recursive Feature Elimination and Principal Component Analysis methods were used to select the most important features that affect the quality of the software code. The selection of characteristics made it possible to select 7–9 characteristics from 21 available. Thanks to this, it was possible to improve the average value of the F-score for all parameters to 0.902-0.905. Data balancing was carried out both by oversampling methods (Random OverSampler, Synthetic Minority Oversampling Technique, KMeansSMOTE, SVMSMOTE), and by undersampling (Random UnderSampler, Near Miss, Tomek Links, Edited Nearest Neighbors) or combined methods (SMOTEENN, SMOTETomek). A method for detecting software defects at the early stages of the life cycle has been developed, which uses a model of software defectivity, which differs from the existing ones by using a limited number of software code metrics that have the greatest impact on defectivity, and consists of a stacking ensemble, which consists of a neural network based on radial basis functions, a recurrent neural network and a long-short-term memory network, which makes it possible to increase the accuracy of software defect prediction. A classifier of the software defect proneness has been created, which is built on an ensemble of unsupervised and supervised machine learning algorithms. The obtained classification accuracy is 0.838, the value of the F-score is 0.909. An information technology for evaluating software reliability and predicting its defectiveness using machine learning methods has been developed. The classification accuracy using the proposed technology is up to 0.920. Based on the created information technology, a prototype of the web service for evaluating and predicting software reliability has b  Product Description popup.authors Izonin Ivan Boyko Natalya Kryvenchuk Yurii Melnykova Natalya Shakhovska Natalya popup.nrat_date 2023-02-06 Close