Abstract
- The recognition of the different phases of human gait is valuable in areas such as rehabilitation and sports. Machine Learning models have been increasingly used for such recognition tasks. However, such models are usually trained on data obtained from participants in strictly controlled environments which—needless to say—might vary quite significantly from the environment in which the models are subsequently employed. Therefore, it is advisable to analyze the confidence of the model’s predictions. To this end, we present an interpretable classifier for gait phase detection. Together with classification reliability estimation tools, classification predictions can be rejected in low confidence scenarios. Our classifier is based on a robust and distance-based Learning Vector Quantization classifier. Finally, we present our approach using a real-world application in gait phase detection, which consists of one learning scenario and two different prediction scenarios.