Determining the content of implementation
Obtain data
Data preprocessing
Method selection
Hyperparameter selection
Model learning
Model evaluation → Go to 3, 4, 5
・ At the time of system release All the data you have may be used as training data
・ If you want to evaluate accuracy Separate training data and test data, and evaluate test data with a model learned only from training data
Because the purpose of supervised learning is to predict unknown data
・ The prediction accuracy is low even for training data.
・ It fits well with the training data, but the prediction accuracy for the test data (unknown data) is low.
The most difficult part of machine learning is how to strike a good balance between underfitting and overfitting.
Recommended Posts