IMU-Based Transportation Mode Recognition using Ensemble Learning and Multisensor Fusion

Transportation mode recognition using mobile sensors has gained importance in urban planning, mobility services, and environmental assessment. However, existing approaches often struggle with generalizing across varied device wearing positions. This paper presents a novel method for position-independent Transportation mode recognition using Inertial Measurement Unit data and ensemble learning. We introduce a comprehensive feature extraction pipeline and an ensemble model combining position-specific Random Forest classifiers with a Gradient Boosting meta-learner. This work advances the field by showcasing the power of multisensor fusion and ensemble learning in overcoming position-dependent limitations in Transportation mode recognition. Experiments on the Sussex-Huawei Locomotion dataset demonstrate our approach’s effectiveness, achieving a 97.49% F1-score using only Inertial Measurement Unit data, outperforming recent methods that incorporated GPS.