Forecasting Project Performance Using a Neural Predictor Model

Abstract

Recently a Project Assessment Indicator (PAI) Model has been applied in the Project Management Field. The PAI Model output or PAI indicates the entire project performance at the current but not at the future time. We use a neural network (NN) to predict the project performance based on past PAI data. The NN model has been tested on monthly PAI and virtual weekly PAI generated using the 2^nd Newton Interpolation Function. Generating weekly PAI using our approach helps to increase the dataset size for training the neural predictor model and allows the project manager to make weekly predictions of project performance. The project manager may then have quantitative estimates of the future performance of the project at intervals shorter than one month. PAI is normally reported on a monthly basis. The NN model performance is evaluated using the Mean Absolute Percantage Error. We achieve a predictive accuracy ranging from 1.21% to 2.87% on the test data of the virtual weekly PAIs. This suggests that depending on the dataset size, project managers could use the NN model to predict future project performance to about 1.21% ∼ 2.87% of the weekly PAI value.