… to analyze sports motions. The first involves biomechanical feature analysis. We adopt a classical pipeline that trains a machine learning model and interprets it using explainable AI techniques. A quantile random forest regressor is employed to emphasize top performance, and the model is interpreted using SHapley Additive exPlanations (SHAP), along with Partial Dependence Plots (PDPs) and Individual Conditional Expectation (ICE) plots. The analysis yields insights consistent with prior literature. However, this approach is limited in its ability to capture interactions between features, which are often critical. To address this, we introduce a novel method to estimate functional interactions, which not only quantifies interaction strength but also characterizes how and where these interactions occur. The second approach focuses on analyzing pose sequences, with an emphasis on interpreting black-box time-series models. We generate counterfactual explanations using a sparse autoencoder-based framework. Experimental results demonstrate that our method produces explanations that are both faithful and robust.