Prediction Project Task Completion Using Supervised Machine Learning Method: A Conceptual Approach

Abstract

Project schedule forecasting is a core enabler of successful project management. Accurate schedule prediction leads to better resource management and ultimately, more value gained from the investment made for the project. The higher the complexity of the project, the higher the importance of having an accurate schedule prediction to minimize the risks associated with the project. The Field X Expansion Project of Company Y provided an excellent case study of the successful pilot implementation of supervised machine learning to predict the completion of the project tasks, which gave more precise results compared to the existing conservative approach. The Field X Expansion Project was designed to increase the total daily production from the gigantic Field X reservoir. The project’s cost was in the multi-billion dollars range, making it one of the highest investments of the decade in the oil and gas industry. Therefore, it is crucial to complete the project on schedule and within the budget to maintain its economic value. However, there were multiple challenges in the project that brought uncertainties and complexities to the schedule prediction, which cannot be solved using the conservative approach, such as the challenges in the project terrain and geography, the weather, and the mobilization of project logistics from around the globe. The conservative approach utilizing the off-the-shelf project management software has attempted to forecast past projects schedule more accurately. With this software, each project task and its estimated duration serve as inputs for the software to calculate the estimated project completion. To the team’s disappointment, the result showed overall schedule accuracy of only 40%. Moreover, using this method, the software can only calculate the estimated completion of the whole project, not the completion of the individual tasks. Although useful, it can still be improved. The software has been able to accumulate historical data from many previous projects utilizing this approach to be used as a data source for further improvement. With the advancement of data science technology and the immense amount of accumulated data from previous projects, there is an opportunity to leverage more advanced analytics methods such as big data analytics and machine learning to predict task completion with higher accuracy. This paper discusses the big data analytics approach to predicting individual project task completion. The method involved pulling the task data from the project management software database and analyzing the impact of various variables and features of the project on the completion of the individual project tasks that ultimately affected the project schedule. The features with the most significant impact were then used as predictors to forecast the completion of each project task. Applying this method to the Field X Expansion Project, the task completion can be predicted with 98.6% accuracy and 90% Receiver Operating Characteristic Area Under Curve (ROC AUC). This result is higher than the baseline accuracy of 40% applying the conservative approach. With the new accuracy, the project quality is improving, thus avoiding the loss of millions of dollars from poor project management.