Task/optimize model hyperparams

[IshantSingh24]

Description

optimize model hyperparams

Semver Changes

Model Architecture

Input (250, 1)
↓
Bidirectional(LSTM(128, return_sequences=True))
↓
Dropout(0.4) + BatchNormalization
↓
Bidirectional(LSTM(64))
↓
Dropout(0.4) + BatchNormalization
↓
Dense(64, ReLU)
↓
Dropout(0.4)
↓
Dense(32, ReLU)
↓
Dropout(0.2)
↓
Dense(num_classes, Softmax)

Training Details:

• Optimizer: Adam (lr = 0.001)
• Loss:Categorical Crossentropy
• Metrics: Accuracy, Precision, Recall
• Batch Size: 256
• Max Epochs: 50
• Callbacks:
  • EarlyStopping: Monitors val_loss (patience=15, restore_best_weights=True)
  • ReduceLROnPlateau: Monitors val_loss (patience=5, factor=0.2)

---

Performance

Metric	Value
Accuracy	~98.5%
Precision	~98.5%
Recall	~98.5%
F1-Score	~98.5%

Issues resolved

Improved lacking areas of the model like F1 score.
Much better prediction of class F, S, V.

Checklist

• [ yes] I have read the Contributing Guidelines.