Recent advancements in machine learning offer promising avenues for analyzing functional magnetic resonance imaging (fMRI) data. Yet, the models’ “black-box” nature hampers their interpretability and explainability. This study explores the application of explainable artificial intelligence (xAI) to enhance the interpretability of fMRI data analysis conducted with Artificial Neural Networks (ANNs). Leveraging insights from xAI, we aim to address these concerns by elucidating the inner workings of ANN models. Our method involves constructing a feedforward backpropagation shallow neural network (SNN) with a simple architecture conducive to explainability. Through training and testing, we evaluate the SNN performance in classifying fMRI data related to the motor paradigm of HCP’s Young Adult database. Additionally, we incorporate pruning and retraining stages to enhance model efficiency and interpretability. The elbow points on the path-weights distribution are introduced as a data-based pruning threshold. Results demonstrate that while pruning initially leads to a decrease in accuracy, subsequent retraining mitigates this effect, yielding a refined network with improved performance. Moreover, SHAP value analysis provides insights into the most influential features driving classification decisions. This study concludes that the elbow-point pruning threshold yields a pruned network that retains more accuracy than the arbitrary methods previously used at the cost of higher model complexity. The retrained network did not recover as much accuracy as in previous studies, but this can be attributed to a poor performance for the LF class. Finally, the SHAP explanation is consistent with path-weight-based explanations, LRP-based explanations, and established neuroscientific knowledge.