Multi-Layer Perceptron
Neural network classifier for complex nonlinear patterns
Multi-Layer Perceptron (MLP) is a feed-forward neural network with one or more hidden layers. It can model highly nonlinear relationships but requires more data and tuning compared to tree-based models.
When to use:
- Complex nonlinear interactions between many features
- Datasets large enough to support neural network training
- When ensemble methods have been exhausted and further accuracy gains are needed
Input: Tabular data with the feature columns defined during training Output: Predicted class label and class probabilities
Model Settings (set during training, used at inference)
Hidden Layer Sizes (default: (100,))
Architecture as a tuple of layer sizes. E.g., (100, 50) creates two hidden layers.
Activation (default: relu)
Activation function for hidden layers. relu is standard; tanh can work better for certain distributions.
Solver (default: adam)
Weight optimization algorithm. adam adapts learning rates automatically; sgd gives more control with manual tuning.
Alpha (default: 0.0001) L2 regularization term. Increase to reduce overfitting on small datasets.
Learning Rate Init (default: 0.001)
Initial learning rate for adam and sgd.
Max Iter (default: 200) Maximum training iterations (epochs).
Early Stopping (default: false) Stops training when validation score stops improving.
Inference Settings
No dedicated inference-time settings. The trained neural network weights are applied at prediction time.