Article Preview
Top1. Introduction
Software reliability modeling is an expertise field of investigation and research in software engineering. Various different schemes have been given which includes time dependent fault detection and correlation model building. The common methodology is to adopt a data source and difference of estimated and observed values are measured to implement the proposed model.
The primary aim of various fault prediction techniques is to address defects presents in software modules as early as possible (Raghuvanshi et al., 2021). The inherent characteristics of software leads to identify bugs exist in the system. The fault prediction models play a major role in success of a software and is considered as one of the important tasks. Because if faults are present in the system that it will definitely impact the quality of the software thereby increases maintenance and development cost. By predicting number of faults in software modules, we can guide software testers to focus on faulty modules first (Ivanov et al., 2018; Nagaraju et al., 2019). To evaluate the impact of software failure prediction limited testing resources might be available as the distribution of faults/defects in software may be extreme at one position and might be nominal in certain cases. This uncertainty effect must be considered while designing a software reliability model. This uncertainty is carefully examined by software developers and project teams that will employ substantial resources, time and effort, in recognizing and fixing defects/faults described by the stakeholders. The defect prediction in early stages of development like coding will help in reducing defect amplification (Rathore & Kumar, 2017).
In this article we propose a Fault Prediction Model to predict software faults and improve software reliability. The main highlights of this work can be concise as follows:
- •
The presented work uses software team assessment and prediction dataset to formulate the prediction model based upon few deterministic attributes such as Team-Member-Count, Average-Coding-Deliverables-Hours-Week, Average-Responses-By-Week, Average-Non-Coding-Deliver- ables-Hours-Week and Average-Commit-Count.
- •
To build the prediction model and to assess its impact some project specific measures are determined which includes professional expertise, team member count and distribution, the complexity of product/process, the process of testing/debugging and organizational policies.
- •
The proposed model is implemented using JIRA software and the results are equated with similar work in literature based upon error rate, accuracy and information criteria.
The presented article is structured into sections. In Section 2, the literature work is explained along with the contextual and impulsive work of this study. Section 3 offers a brief outline of NHPPM and the conventions followed while designing the proposed fault prediction model. In Section 4, we discuss the methodology along with dataset, system setup and the benchmarks for models' performance comparison. In Section 5, we present results and gives the graphical comparison of related models. This section includes the performance evaluation study and predictive accuracy of the presented work with some prevailing models. Finally, the presented work is concluded along with the listing of future directions in Section 6.
Top2. Background
This section presents the related work carried out in the field of software reliability and prediction analysis. A numerous research work is carried out in this field where each author proposed a different strategy to countermeasure the issues exist with fault prediction in software engineering.
Gao and Khoshgoftaar (2017) in their research provide an experimental study for fault prediction in software. The analysis is carried out upon categorical Poisson regression model, negative binomial regression model and hurdle count models. Various quantitative attributes have been identified and assessed for prediction of software quality. The survey results into evaluation of different count models on the basis of a One-way Anova and Tukey's method. A machine learning based model is suggested by the authors of Wang & Zhang, (2018). A deep neural network model is suggested for improved results in software reliability. With the approach given in Li & Pham, (2017) the authors presented an error generation and fault removal model for software reliability. The aim is to assess fault removal efficiency including the introduction and testing coverage.