AAI_2025_Capstone_Chronicles_Combined

Figure 5.6 MNIST Confusion Matrix Differences

While the ensemble model outperforms the single PyTorch model in terms of accuracy, it requires significantly more time to train. Nevertheless, this experiment demonstrates the versatility of the TurbaNet library—it can be leveraged either to reduce training time in certain scenarios or to enhance model performance in others, depending on the use case.

Table 5.1: Model Performance Metrics for MNIST Dataset Model Accuracy

Training Time (s)

Precision

Recall

F1-Score

43.62

PyTorch

92.9%

92.8%

92.8%

92.8%

Turba (Avg)

124.79

93.8%

93.7%

93.7%

93.7%

94.4%

94.4%

94.4%

94.4%

Turba (Ens)

124.79

142