APPLICATION OF EVOLUTIONARY TECHNIQUES FOR SOFTWARE EFFORT ESTIMATION

Abstract

Software effort estimation is a very difficult task carried out by software project managers as very little information is available in the early phases of software development. The information that we are collecting about various attributes of software needs to be subjective which otherwise can lead to uncertainity. Inaccurate software effort estimation can be disastrous. Both underestimation and over estimation can lead to schedule overruns and incorrect estimation of budget for software development. Software effort estimation is a very crucial activity for project control, quality control and success of any software project. Software effort estimation fall under the categories of expert judgement, algorithmic and machine learning techniques. We have tried to analyse the performance of evolutionary techniques for software effort estimation. For this purpose various datasets with different properties have been collected. After that various evolutionary algorithms like FRSBM-R, GFS-SAP-Sym-R, GFS-GAP-Sym-R, NNEP-R, GANN-R, GFS-GP-R, GFS-GSP-R, GFSRB- MF-R, CART-R, Linear_LMS-R, NU_SVR-R, EPSILON_SVR-R etc have been used. Performance is measured in terms of various accuracy measures like MMRE, MRE, PRED(25), PRED(50) and PRED(75). Results of our research have shown that evolutionary algorithms give more accurate results for software effort estimation as compared to traditional methods of software effort estimation. Moreover the comparison of different evolutionary algorithms is done to find which evolutionary learning algorithm is better for which situation.