Figure 1.

Comprehensive evaluation of a hybrid neural network model for classifying five types of activities from accelerometer data. Top-left: Confusion matrix displaying the recall of model classifications for activities (‘other’, ‘eating’, ‘exercise’, ‘medication’, and ‘smoking’). Top-right: Scatter plot of the macro F₁ score versus average network confidence, with a Pearson correlation coefficient (r = 0.71) indicated by the red line. Bottom-left: Box plots showing the distribution of confidence, F₁ score, precision, and recall metrics across cross-validation folds, illustrating performance consistency and variability. Bottom-right: Curve illustrating the relationship between the confidence threshold and the average F₁ score, demonstrating how model performance optimizes at higher thresholds. These panels collectively highlight the model’s effectiveness and potential utility in real-world applications for health monitoring and behavior analysis.