Feasibility test of activity index summary metric in human hand activity recognition

Activity monitoring is a technique for assessing the physical activity that person undertakes over some time. Index (AI) metric summarizes raw measurements from tri-axial accelerometers, often used measuring activity. Our research compared different groups and hand usage [1]. We also tested this as classification feature, how data acquisition segmentation parameter configurations influence accuracy. Data was done with previously developed system includes smartwatch on each wrist smartphone placed in subject?s pocket; accelerometers analysis. calculated labeled segments ANOVA1 statistical test Bonferroni correction. Significant differences were found between cases of (left, right, none, both). In next analysis phase, feature three supervised machine learning algorithms-Support Vector Machine, k-Nearest Neighbors, Random Forest. The best accuracy (measured by F1 score) classifying achieved using Forest algorithm, 50 Hz sampling frequency, window 10 s without overlap AI calculation, it 97%. On other hand, had low accuracy, which indicated specific group can?t be identified only one simple feature.