Piecewise Congruence Regressed Indexive Extreme Learning classifier for Software Fault Prediction

Abstract

Early detection of software faults is significant task in the software development process. With the development of defect prediction technology, software with defects negatively impacts on operational costs and finally affects client satisfaction. Therefore, early defect prediction by the project managers is critical and challenging task. Different approaches were developed to predict software faults. However two essential factors are major issues such as timely and accurate detection. Therefore, a novel technique is required for accurate and timely software fault production. In this paper, a Piecewise Congruence Regressed Indexive Extreme Learning Classifier (PRILEC) is introduced for accurate software fault prediction with minimum time. The PRILEC technique consists of two major processes namely feature selection or software metric selection from the input dataset and the classification. In the first process, feature selection or software metric selection process is carried out by using congruence correlative piecewise regression. First, the numbers of features or metrics are collected from the dataset. Then the correlation between the features is measured using congruence coefficient and identifies the more relevant features. With the selected features, the classification is performed by applying statistical indexive levenberg extreme learning classifier for fault prediction with higher accuracy. In the extreme learning classifier, the testing and raining data analysis is performed with the help of Camargo's statistical index. Then the Hardlimit activation function is applied to identify the defective or non-defective software code. Finally, the Levenberg–Marquardt algorithm is applied to minimize the least square problem and obtain the final better classification results at output layer. In this way, software fault prediction is carried out with higher accuracy and minimum time. Experimental assessment of the proposed technique is carried out with respect to fault prediction accuracy, precision, recall, F-measure, prediction, specificity and time complexity with a different number of data. The quantitatively discussed results indicate that the performance of proposed technique increases data accuracy of software fault prediction with a higher minimum time as well as space complexity than the conventional method.