Trajectory Classification Using Feature Selection by Genetic Algorithm

Abstract

Trajectory classification helps in understanding the behavior of objects being monitored. The raw trajectories may not yield satisfactory classification results. Therefore, features are extracted from raw trajectories to improve classification results. All the extracted features may not be useful for classification. Hence, an automatic selection scheme is essential to find optimal features from the pool of handcrafted features. This paper uses a genetic framework to choose the optimal set of features for trajectory classification purpose. Seven features costing 18 dimensions have been extracted from raw trajectories. Next, Genetic Algorithm (GA) has been used to find the optimal set of features from them. The binary encoding scheme has been used in GA. The 7-bit long chromosomes have been coded in this work. Bits of chromosomes represent trajectory features to be used in classification. Finally, trajectories have been classified using optimal features. Trajectory classification has been done using Random Forest (RF) based classifier and compared with Support Vector Machine (SVM). The results are evaluated using three trajectory datasets, namely I5, LabOmni2, and T15. The classification rates of 99.87%, 93.32%, and 90.58% have been recorded for datasets I5, LabOmni2, and T15, respectively. © 2020, Springer Nature Singapore Pte Ltd.