DEVELOPMENT OF A PROJECT PROGRESS PREDICTION MODEL FOR THE MARKETING AND PROJECT MANAGEMENT UNIT OF TELKOM PROPERTY AREA III (WEST JAVA) USING DYNAMIC TIME WARPING, LOGISTIC REGRESSION, AND SUPPORT VECTOR REGRESSION

• Main Author: AKBAR MAULANA, MUHAMMAD
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/42752
• Institution: Institut Teknologi Bandung

PT Graha Sarana Duta (Telkom Property) is a subsidiary of PT Telekomunikasi Indonesia (Telkom) which operates in the property business. Its main tasks include managing and maintaining Telkom’s various property assets, which are spread throughout Indonesia. The company commonly utilizes the services of various subcontractors in completing its projects. Hence, the company’s main responsibilities include managing, supervising, and controlling said projects to ensure that they are completed on time. To prevent projects from finishing late, Telkom Property needs to be able to detect and mitigate potentially late projects far ahead of their deadline. However, considering the volume and variety of the company’s current projects, this is not an easy task to do. To solve this problem, a multi-step forecasting model for project progress is proposed, which will be used to forecast project progress up until its deadline. With this model, company staff will be able to determine which projects are most likely to finish late and focus their attention and efforts on those projects, which will allow them to monitor the various projects much more effectively. In this final project, before constructing a forecasting model, the company’s various projects are first clustered according to their S-curve shape. To accomplish this, a hierarchical clustering procedure with Dynamic Time Warping (DTW) as a similarity measure was utilized. Afterward, a logistic regression model was constructed to classify new observations into each cluster. Finally, for each cluster, a Support Vector Regression (SVR) model was built to predict project progress. As a result of the clustering process, two clusters were generated. Cluster 1 contains projects which could mostly be completed within one week, and Cluster 2 contains projects where project completion progressed more gradually. Based on the logistic regression coefficients, Cluster 1 mainly contains projects concerning pavement and electrical works, as well as other projects with relatively few activities and low contract values. Cluster 2 mainly contains projects with higher activity counts and contract values. Finally, one SVR model was constructed for each cluster. The two-step test-set iterated forecasting error observed for the Cluster 1 model is 0.0218 and the four-step test-set iterated forecasting error observed for the Cluster 2 model is 0.1104.