Effort estimation models using evolutionary learning algorithms for software development

Software effort estimation is a complicated task being carried out by software developers as very little information is available to them in the early phases of software development. The information collected about various attributes of software needs to be subjective, which otherwise can lead to uncertainty. Inaccurate software effort estimation can be disastrous as both underestimation and overestimation may result in schedule overruns and incorrect estimation of budget. This paper focuses on the comparative study of various non-algorithmic techniques used for estimating the software effort by empirical evaluation of five different evolutionary learning algorithms. The accuracy of these algorithms is found out and the behavior of these algorithms is analyzed with respect to the size and the type of data. All the five techniques are applied on three different datasets and various paramenters such as MMRE, PRED(25), PRED(50), PRED(75) are calculated. The proposed results are compared to other machine learning methods like SVR, ANFIS etc. The results show that evolutionary learning algorithms give more accurate results than machine learning algorithms.