Physical Exercise Prediction for Healthcare routines over Multi-Sensory Ontology

Madiha Javeed ¹ Mohammad Shorfuzzaman ² Touseef Sadiq ^3* Hameedur Rahman ^1* Asaad Algarni ⁴ Ahmad Jalal ¹

• ¹ Air University, Islamabad, Pakistan
• ² Taif University, Ta'if, Saudi Arabia
• ³ University of Agder, Kristiansand, Vest-Agder, Norway
• ⁴ Northern Border University, Arar, Northern Borders, Saudi Arabia

Physical action prediction or exercise recognition is an important factor for the healthcare industry around the globe. It is prevalent for improved human lifestyle outcomes. This research brings an innovation for physical action prediction using multimodal methodology and machine learning techniques, where we integrate novel descriptor extraction methods to improve the predictive power and ability in healthcare domain.Methods: This article presents a unique model for processing physical activities-based data, including cameras-based captures, wearable sensors-based signals, and ambient-based gestures. A machine learning-related method named the hidden Markov model has been incorporated to speed up the prediction of physical action. Additionally, we have proposed unique methods to extract the descriptors from multiple types of health data. A few experiments over benchmarked datasets have been conducted for data training and validation. The physical exercise prediction system has been assessed using multiple evaluation criteria, such as, accuracy, precision, recall, and F1-score.Our system outperformed the conventional methods, achieving accuracy rates of 81.67% and 80.00% over Berkeley-MHAD and LARa datasets, respectively. The precision and recall metrics demonstrated the system's ability to recognize the exercises performed. The wide-ranging F1-score has also shown outstanding performance by obtaining better results.The proposed system reveals the prospects of multisensory devices and machine learning application for physical action prediction. The consistent efficiency of our system across different datasets has offered new possibilities for the detection and interpolation of the unwanted event for quality lifestyle and good healthcare. This system promises to provide significant application projections and considerable progressions in healthcare domain.