Aiming at the problem that fatigue driving is difficult to be detected accurately and the interpretability of the detection model is low, an interpretable sparrow search algorithm optimized random forest model (SSA-RFC-SHAP) was proposed for driving fatigue detection. The driver's pulse wave signal was used as the data source, the heart rate variability analysis was performed and the feature indicators were extracted. The Pearson's correlation test and chi-square independence test were used to screen out the feature indicators for the determination of driving fatigue. The Random Forest classifier was optimized by the sparrow algorithm, and the driving fatigue three-classification detection model was established. And finally, Shapley's additive explanation algorithm was used for the interpretation of the model detection results. Finally, Shapley's additive explanation algorithm was used to analyze the interpretability of the model detection results. The results show that: the proposed SSA-RFC-SHAP model has high accuracy, and in the driving fatigue three-classification detection task, the accuracy, precision, recall, and F₁ reach 90.52%, 90.34%, 90.16%, and 90.24%, respectively, which is higher than that of RFC, BiLSTM, CNN-LSTM, and Gradient Boosting model. Interpretable analysis from the perspective of the model, the influence of each feature on the model prediction and the specific decision-making process of the model are obtained, which significantly improves the interpretability of the model, in which MeanHR has a negative correlation with the fatigue state, and MedianNN and LF have a positive correlation with the fatigue state. It can be seen that the proposed SSA-RFC-SHAP driving fatigue detection model can provide scientific guidance for driving fatigue warning.