Table 1 Performance comparison of classification models on complex sequential data with cubic (class 1) vs. linear (class 0) polynomial trends and 97% noise features. Metrics represent means over 10\(\times\)5 repeated cross-validation (T=1,000 timesteps, m=100 features). LEW-RF: Legendre Energy-Weighted Random Forest; RF: Standard Random Forest; BiLSTM: Bidirectional LSTM; SVM: Support Vector Machine; LR: Logistic Regression; DT: Decision Tree; GBM: Gradient Boosting Machine.
Model | Accuracy | Precision | Recall | F1 | AUC | Time (seconds) |
|---|---|---|---|---|---|---|
LEW-RF | 81.2% | 84.2% | 76.5% | 79.9% | 86.4% | 0.68 |
RF | 75.9% | 96.9% | 53.1% | 68.3% | 83.5% | 0.78 |
LSTM | 58.7% | 67.4% | 43.7% | 51.4% | 60.6% | 61.41 |
BiLSTM | 73.1% | 89.1% | 53.1% | 65.9% | 74.4% | 85.94 |
SVM | 57.1% | 57.4% | 52.6% | 54.7% | 61.3% | 1.13 |
LR | 49.2% | 48.7% | 48.5% | 48.4% | 49.1% | 0.05 |
DT | 77.3% | 81.3% | 70.6% | 75.2% | 82.4% | 0.20 |
GBM | 77.3% | 83.0% | 68.4% | 74.7% | 84.9% | 0.39 |
