Software Effort Estimation Using Logarithmic Fuzzy Preference Programming and Least Squares Support Vector Machines

Abstract

Purpose: Effort Estimation is a process by which one can predict the development time and cost to develop a software process or product. Many approaches have been tried to predict this probabilistic process accurately, but no single technique has been consistently successful. There have been many studies on software effort estimation using Fuzzy or Machine Learning. For this reason, this study aims to combine Fuzzy and Machine Learning and get better results.Methods: Various methods and combinations have been carried out in previous research, this research tries to combine Fuzzy and Machine Learning methods, namely Logarithmic Fuzzy Preference Programming (LFPP) and Least Squares Support Vector Machines Machine (LSSVM). LFPP is used to recalculate the cost driver weights and generate Effort Adjustment Point (EAP). The EAP and Lines of Code values are then entered as input for LSSVM. The output results are then measured using the Mean Magnitude of Relative Error (MMRE) and Root-Mean-Square Error (RMSE). In this study, COCOMO and NASA datasets were used.Result: The results obtained are MMRE of 0.015019 and RMSE of 1.703092 on the COCOMO dataset, while on the NASA dataset the results of MMRE are 0.007324 and RMSE are 6.037986. Then 100% of the prediction results meet the 1% range of actual effort on the COCOMO dataset, while on the NASA dataset, the results show that 89,475 meet the 1% range of actual effort and 100% meet the 5% range of actual effort. The results of this study also show a better level of accuracy than using the COCOMO Intermediate method.Novelty: This study uses a combination of LFPP and LSSVM, which is an improvement from previous studies that used a combination of FAHP and LSSVM. The method used is also different where LFPP produces better output than FAHP and all data in the dataset is used for training and testing, whereas in previous research it only used a small part of the data.