Effect of Dimensionality Reduction on Prediction Accuracy of Effort of Agile Projects Using Principal Component Analysis

Abstract

Abstract Agile framework for software development has received a lot of recognition in software industry in previous years as it focuses on rapid incremental delivery, lower risk and customer satisfaction. At early stages of development, the effort must be predicted so that the project is completed successfully within the time and cost deadlines. In recent years, various researchers have done study in this area and it is observed that the prediction of effort faces a problem of large dimension of features. Hence the prediction accuracy may be increased by reducing the dimensions of the features. In this paper, PCA has been used for reduction of feature dimensions for effort estimation. PCA identifies the key attributes by reducing the dimensions of the attribute which are those having highest correlation with the effort. The methodology shows the effect of PCA on the original dataset and the results are observed by applying various machine learning techniques pre and post PCA. The comparison metrics used are Mean Magnitude relative Error (MMRE), Root Mean Square Error (RMSE), and Prediction Accuracy (PRED (25)). The decreased values of errors and increased value of accuracy shows the better model accuracy when PCA is applied on the dataset. All the computations and implementations in this paper are done using Python on Scikit-learn library.