Machine learning: avoid data leakage
Features are inputs and labels are the outcome you want to predict. Start with a clear objective and a baseline. Identify features available at prediction time.
Machine learning: avoid data leakage
Features and labels
Features are inputs and labels are the outcome you want to predict. Start with a clear objective and a baseline. Identify features available at prediction time.
Training and evaluation
Separate training data from evaluation data. Select a metric that reflects the cost of mistakes. Accuracy can mislead on imbalanced classes.
Overfitting and leakage
Overfitting means learning training-specific patterns that do not generalize. Data leakage gives the model information it would not have in use. Keep preprocessing fitted only on training data.
Your next step
Try the exercises with a small dataset. Write down your assumptions, check the results against the source, and explain one thing you would improve.