Validation of Existing Software Effort Estimation Techniques in Context with Mobile Software Applications

摘要

In the current generation of information technology, mobile applications (apps) have become an essential and momentous source to publicize the information across the world. Academia, industries and other organizations have preferred mobile apps rather than classical software. Mobile apps are different from classical software and popularity, adaptability of mobile apps is more with wide range use. The growth of mobile apps across various fields has shown a big challenge for mobile app development industries to deliver apps on time and budget with desired accuracy and performance. Planning of mobile-based projects is a very complex task for the software industry, especially estimation of effort, time and cost for development of mobile apps. There are various literature, method, and model available in the field of classical software but mobile apps are different from classical software by their nature. It has also observed that the selection of input data is also affecting the accuracy of prediction. There is lack of calibrated model and method that administer the immense scope of determination in development of effort estimation for mobile apps. In this paper, various existing techniques of effort estimation have applied on software analytics for mobile apps (SAMOA) dataset for better analysis of suitable estimation technique that fits for mobile App development. The aim of this paper is twofold-(i) to explore the performance of variously established estimation technique on mobile app development (SAMOA dataset). (ii) Analysis of experimental results and, suggesting the best technique for the distinguished mobile app development scenario. The work is carried out adopting four techniques namely multiple linear regressions, Multi-Layer Perceptron Neural Network (MLP-NN), Genetic Algorithm (GA) and Naive forecasting approach. The results have compared with these statistical models. Among all techniques, the experimental results have presented that the GA was outperforming among four effort estimation techniques. Mobile app effort estimation models have built using four-estimation technique using SAMOA dataset. In addition, we investigated and compared various techniques namely MLP, MLP-NN, GA and Naive forecasting approach. Upon construction, accuracy measures MMRE, MRE, PRED(25) represented promising outcomes for mobile apps used in the effort estimation model construction and validation of the process. The analysis presented that GA provided better performance rather than another approach.