Figure 5.

Confusion matrices on different evaluation metrics for DeepConvLSTM. Numbers below 0.05 are left out. The color intensity represents the degree of similarity between predicted and actual values: darker colors along the diagonal represent accurate predictions (values closer to 1), while lighter colors off the diagonal indicate misclassifications (values closer to 0). The color intensity serves as a visual cue, emphasizing the model’s performance in correctly predicting different classes. (a) k-fold cross-validation. (b) Leave-recordings-out cross-validation. Legend: 1 = assist in getting dressed; 2 = full body washing in bed; 3 = feeding; 4 = make the bed; 5 = clean up; 6 = skin care; 7 = push the wheelchair; 8 = wheelchair transfer; 9 = comb hair; 10 = wipe up; 11 = prepare medication; 12 = serve food; 13 = documentation.