Missing Data Imputation using K-Nearest Neighbour for Software Project Effort Prediction

Abstract

The accurate of software development effort prediction plays an important role to estimate how much effort should be prepared during the works of a software project so that it can be completed on time and budget. Achieving good prediction accuracy is rely on the quality of data set. Unfortunately, missing data is one of big problem regards to the software effort data set, beside imbalance, noisy and irrelevant problem. Low quality of data set would decrease the performance of prediction model. This study aims to investigating the accuracy of software effort prediction with missing data set by using KNN missing data imputation and List Wise Deletion (LWD) techniques. It was continued by applying stepwise regression with backward elimination for feature selection and implementing two effort prediction methods of Multiple Linear Regression (MLR) and Analogy. The result shows that missing data imputation using KNN and listwise deletion with multiple linear regression approach outperforms the Analogy approach significantly (p>0.05).