Software Effort Estimation using ISBSG Data

Abstract

Effort Estimation is a process of predicting the amount of effort required for the development of a specific Software Project. Effort estimation can be divided into two parts mainly algorithmic and non-algorithmic. Algorithmic estimation was done in the starting years of effort estimation. With the Increase in complexity of software activities effort estimation also became a complex task and thus to overcome the limitations of algorithmic approaches non-algorithmic approaches are used. Although a lot of research is done but there is No validation for superiority of one model over other. The main objective of this work is to study the usability of random forest in effort estimation and select important features using random forest feature selection. In this work, different machine learning models are built on ISBSG release 2019 dataset. The results of different models are compared with each other using two accuracy measures. In this Study, we found out the most influential features in effort estimation then three models are built on the selected features. Then, the performances of model are compared to find the best fitting model for this dataset based on two different measures, MMRE and PRED (0.25). Results obtained from the experiment shows RF obtained PRED (0.25) score of 95% with best MMRE of 0.08. The trained RF model shows the important features to be Function Size and Project elapsed time.