End-to-End Machine Learning Pipeline

ML_outline.md

End-to-End Machine Learning Pipeline

Outline of the process of for developing a ML project.

• Explore and clean the data.
  • Clean continuous features.
  • Clean categorical features.
• Split the data three ways.
  • Training - 60%
  • Validation - 20%
  • Testing - 20%
• Fit an inital model and evaluate using fivefold cross-validation of the training data.
  • A Random Forest classifier is used for decision trees.
• Tune hyperparameters.
  • For a Random Forest, tune the number of estinators (decisions) and the max depth.
  • A grid search can be used.
• Evaluate some models using the validation set.
  • Look at performance metrics beyond just accuracy.
  • Use the three best models based on the training set.
  • Predict on the validation set and look for the best accuracy, precision, and recall based on the context of the data.
• Finally, select the best model and evaluate on the testing test.
  • Does the test set match the model we picked?
  • Deploy the best model.