Comparison of SVM and Random Forest for Multimodal Hand Gesture Recognition

Machine learning has been implemented to detect and identify sign language hand gestures (letters and numbers) using the Photoplethysmography (PPG) and impedance. The sensor data was collected from commercially available wristbands and evaluation boards. Two supervised machine learning algorithms: Random Forest and Support Vector Machine were used for categorizing the data points. The accuracy for each of the algorithms for the letters and numbers was within the range of either 70-75% or 20-28% for the PPG data. The accuracy was between 33-40% and 55-66% using Random Forest. The accuracy was between 61-76% to 80-90% for when using Support Vector Machine for impedance data. Between the two algorithms, the Random forest approach has better consistency of 70-75% with PPG data whereas the Support Vector Machine approach is more consistent for impedance data with 61-76% accuracy to detect the hand gestures.