Overfitting
When a model learns training data too well, including noise and outliers, leading to poor performance on new data.
In-Depth Explanation
Overfitting occurs when a machine learning model learns the training data too precisely, memorising specific examples rather than learning general patterns. This results in excellent training performance but poor generalisation to new data.
Signs of overfitting:
- Training accuracy much higher than validation accuracy
- Performance degrades on new data
- Model makes confident but wrong predictions
- High variance in predictions
Causes of overfitting:
- Insufficient data: Not enough examples to learn patterns
- Too complex model: More capacity than needed
- Training too long: Model starts memorising
- Noisy data: Learning noise as signal
Prevention techniques:
- More data: Expand training set
- Regularisation: L1/L2 penalties, dropout
- Early stopping: Stop when validation peaks
- Cross-validation: Test on multiple data splits
- Simpler model: Reduce capacity
- Data augmentation: Create training variations
Business Context
Overfitting means your AI works great on test data but fails in production. Proper evaluation and validation prevent this costly problem.
How Clever Ops Uses This
We use proper validation techniques to ensure AI solutions for Australian businesses generalise well to real-world data, not just test scenarios.
Example Use Case
"A model memorises training examples perfectly but can't generalise to new customer queries - catching this requires proper validation."