Alternative Gait-Based Identification by Advanced Machine Learning With Compact Data Learning

Gait recognition as a distinctive biometric modality leverages unique human walking patterns for identification, contributing significantly to public safety, surveillance systems, and identity verification applications. However, traditional methods often face challenges such as high computational cost and data redundancy. A novel Compact Data Learning (CDL) framework is introduced to enhance gait-based recognition systems through dataset size reduction without compromising performance. Integration of feature and sample reduction techniques within CDL generates compact datasets before training, resulting in lowered computational demands and elevated operational efficiency. Four machine learning (ML) models including Random Forest (RF) and Long Short-Term Memory (LSTM) models undergo evaluation on a public gait dataset. Experimental outcomes reveal that CDL achieves 42% dataset size reduction while preserving high classification accuracy. The RF model demonstrates optimal trade-off between accuracy (94.54%) and training duration (7.88 seconds), rendering it most appropriate for deployment. Results demonstrate the effectiveness of CDL in enhancing scalability and efficiency of biometric systems within resource-constrained environments.