Step Length Classification Using Decision Tree Based on IMU Sensors and Body Height

The ability to walk independently contributes significantly to the protective health benefits and complications reduction in stroke rehabilitation. Taskoriented training involves the repetition of functional activities to improve walking ability and is considered a critical strategy. Human gait analysis, especially step length measurement, plays an important role in monitoring mobility and rehabilitation progress, though practical application in everyday life can seldom be achieved by means of conventional methods. This paper presents the confirmation that step length classification is achievable using Inertial Measurement Unit sensors. Feature extraction, filtering, and preprocessing were done on the data from the accelerometer to enhance the accuracy of the classification model. A decision tree algorithm was used for its simplicity and interpretability to classify the step lengths into short (22 cm), medium (33 cm), and long (44 cm). Three confusion matrices are presented, showing the performance of three different models in classifying data into classes 0, 1, and 2. For example, the first model is most accurate for class 0, at an accuracy of 34.88%, but has difficulties classifying classes 1 and 2 at an accuracy of 12.40% and 15.50%, with considerable misclassifications each. Then, the second model yields higher results for both classes 1 and 2 with accuracy of 19.38% each, though it still shows high misclassifications. The third model is the most accurate and balanced; class 0 is classified at 24.07%, class 1 at 25.93%, and class 2 at 27.16%, with low misclassification rates. On the whole, this model is outstanding since it offers the most reliable classification results of all the models.