Figure 3
The chart is a grouped vertical bar graph with five model categories along the horizontal axis (Modes) labeled from left to right as “C N N (Res Net 18)”, “L S T M”, “C N N plus L S T M”, “Vision Transformer”, and “H I P M”. The vertical axis is labeled “Metrics” and ranges from 0 percent to 100 percent with an interval of 10 percent. Each group contains four colored bars representing “Precision” (pink), “Recall” (orange), “F 1 score” (blue), and “A U C” (green), with a legend in matching colors along the bottom. The numeric values are printed on top of each bar. For C N N (Res Net 18): the precision is 80.33 percent, the recall is 77.15 percent, the F 1 score is 75.89 percent, and the A U C is 92.61 percent. For L S T M: the precision is 47.79 percent, the recall is 59.74 percent, the F 1 score is 51.92 percent, and the A U C is 77.15 percent. For C N N plus L S T M: the precision is 88.22 percent, the recall is 82.77 percent, the F 1 score is 84.15 percent, and the A U C has 91.11 percent. For vision transformer: the precision is 68.25 percent, the recall is 70.41 percent, the F 1 score is 68.39 percent, and the A U C is 84.55 percent. For H I P M: the precision is 91.52 percent, the recall is 91.20 percent, the F 1 score is 91.20 percent, and the A U C has 95.36 percent.Performance comparison of different deep learning models – CNN, LSTM, CNN + LSTM, Vision Transformer and the proposed HIMP model – evaluated using precision, recall, F1 score and AUC. The bar chart visually illustrates the effectiveness of each model, with the HIMP model demonstrating superior performance across all metrics, achieving the highest F1 score (91.20%) and AUC (95.36%). These results highlight the advantage of the HIMP model over traditional architectures in this task. Source: Authors’ own work