FSA-Based Model Outperforms Non-FSA Models in Accuracy Across Multiple Datasets

This research examines the performance of FSA-based models compared to non-FSA models, including Boruta-based models, across ten datasets. As shown in Table 11, FSA-based models demonstrate significantly higher accuracy than non-FSA models, including Boruta-based models, on three datasets (ETH, NKK, GLD). This indicates that FSA contributes significantly to improving accuracy on these specific datasets. Furthermore, FSA-based models consistently rank among the top performing models across all ten datasets. In contrast, a Boruta-based method only achieves top performance on four datasets. Notably, these four datasets also have plain models (one of LR, ANN CNN, and LSTM) that rank among the top performers, suggesting that Boruta does not offer a significant advantage in these cases.