Abstract:

Traditional decision tree algorithms are susceptible to bias when certain classes dominate the dataset and prone to overfitting, particularly if they are not pruned. Previous studies have shown that combining several models can mitigate these issues by improving predictive accuracy and robustness. In this study, we propose a novel approach to address these challenges by constructing multiple selective decision trees using the entirety of the input dataset and employing a majority voting scheme for output forecasting. Our method outperforms competing algorithms, including KNN, Decision Trees, Random Forest, Bagging, XGB, Gradient Boost, and Extra Trees, achieving superior accuracy in five out of ten datasets. This practical exploration highlights the effectiveness of our approach in enhancing decision tree performance across diverse datasets.

Speaker Bio :

Mr. Tasnemul Hasan Nehal is a Master Student in the Control and Instrumentation Engineering Department at King Fahd University of Petroleum and Minerals, Saudi Arabia, specializing in Systems and Control Engineering. His research focuses on mathematical modeling, nonlinear control, sensor fusion, and real-time system optimization, blending both theoretical advancements and experimental validation to improve control system reliability. He also has hands-on experience in developing novel machine learning algorithms, computer vision, smart materials, and embedded system design. Currently, his work centers on integrating hardware-in-the-loop (HIL) systems to advance data-driven control strategies and autonomous decision-making.