Classifying Sign Language Gestures using Decision Trees: A Comparison of sEMG and IMU Sensor Data

The use of machine learning technologies for the identification of sign language has gained popularity recently. This arises from the recognition of sign language as a valuable means of communication specifically designed for individuals who are mute or hearing-impaired. In order to build an optimised model based on sEMG, accelerometer, gyro, and magnetometer data, this research article compares decision tree classifiers, notably J48, Random tree, REPTree, and Random forest. This data is collected through the Myo arm band worn on both forearms of the user. The experiment is designed using the open-source Waikato Environment for Knowledge Analysis (WEKA) framework. To evaluate the effectiveness of the four algorithms, three attribute selection techniques information gain-based, correlation-based, and learner-based feature selection were used. The trial results showed that, among the investigated algorithms, the Random Forest method had the highest accuracy, measuring 97.9472%.