A vision-based hybrid ensemble learning approach for classification of gait disorders

• Published in: Multimedia tools and applications Vol. 84; no. 17; pp. 17597 - 17644
• Main Authors: Kour, Navleen, Gupta, Sunanda, Arora, Sakshi
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.05.2025; Springer Nature B.V
• Subjects: Computer Communication Networks; Computer Science; Computer vision; Data Structures and Information Theory; Decision trees; Ensemble learning; Gait; Multimedia Information Systems; Parkinson's disease; Particle swarm optimization; Performance evaluation; Performance measurement; Segmentation; Special Purpose and Application-Based Systems; Track 2: Medical Applications of Multimedia; Vision-based; Knee osteoarthritis; Segmentation; Ensemble learning; Gait analysis; Parkinson’s disease
• ISSN: 1573-7721; 1380-7501; 1573-7721
• DOI: 10.1007/s11042-024-19673-z

Computer vision-based (VB) gait analysis has become the popular platform for detecting Knee Osteoarthritis (KOA) and Parkinson’s disease (PD). The scrutinization of the literature revealed the heavy usage of sensor and markerless platforms but involved certain issues such as exposure to harmful radiations, wearing discomfort, a requirement of background, etc. Further, some aspects are lacking in the previous studies including the exploration of the marker-based (MB) approach, experimentation on disease severity levels using enhanced learning techniques, comparison of abnormal and normal (NM) gait, etc. Therefore, this research aims to predict the pathological and NM gait based on the marker-based (MB) VB platform. In this paper, first, a VB gait dataset is used namely “KOA-PD-NM” which includes three stages: KOA i.e. Early (EL), Moderate (MD), Severe (SV); PD i.e. Mild (ML), MD, SV, and NM subjects, thus, forming a total of seven labels. Then, an improved technique namely Color Segmentation based Fractional Order Darwinian Particle Swarm Optimization (CS-FODPSO) is employed to segment the region of interest (ROI). Next, a hybrid ensemble using k-nearest neighbor (KNN), Decision tree (DT), and Naive Bayes (NB) is proposed to predict the gait patterns of the considered groups. The efficiency of the proposed methodology is evaluated based on performance metrics. The evaluation results achieved provided the highest results using the presented segmentation and hybrid ensemble approaches within less time in comparison to other techniques as well as state-of-the-art. Graphical abstract