Reasons for overfitting: irregular data, small amount of data, complex algorithm
solve:
Normalize data, handle missing values, increase data size, sample, add noise
Regularization, control model complexity
early stoping
Adjust learning rate
dropout
Reason for underfitting: the model complexity is too low, and the number of features is small
solve:
Add new features (polynomial features, high-order features)
Reduce the regularization parameter
Use nonlinear models (SVM, decision trees)