Definition of the ML workflow. The machine learning workflow is composed by the typical phases implemented during a machine learning project (see Figure 4):
(1) Use case conception and problem formulation: This step consists of the definition of the valuable information to get out of the model.
(2) Data preparation: The potential usefulness and accuracy of the model will depend on the quality of the data gathered during this step. Tasks to be performed will include identifying sources of data, aggregating, and cleaning them into a single dataset.
(3) Features engineering and selection: This step covers the understanding and enhancement of the data. Adding new features to a dataset is a key step in the machine learning process.
(4) Dataset split: This phase involves breaking processed data into three datasets—training, validating, and testing:
1. Training dataset—used to initially train the algorithm and teach it how to process information.
2. Validation dataset—used to estimate the accuracy of the model during the training process.
3. Testing dataset—used to compare the accuracy and performance of the models. This set is also meant to expose any issues that might have appeared during training.
(5) Model building and tuning: Choose the best model suitable to solve the user's problem and define their parameters.
(6) Model training and testing: This step involves feeding the training dataset to the model. It will then be able to learn appropriate parameters from the features.
