Table 12 Feature importance score (normalized) across models.

From: PatternFusion: a hybrid model for pattern recognition in time-series data using ensemble learning

Feature category

LSTM

CNN

Transformer

PatternFusion

Spectral power

0.40

0.35

0.33

0.90

Temporal dynamics

0.60

0.57

0.35

0.80

Shape similarity

0.63

0.59

0.59

0.95

Volatility

0.39

0.48

0.44

0.70

Autocorrelation

0.31

0.33

0.44

0.85

Cross-correlation

0.32

0.52

0.39

0.75

Back to article page