Starting from tabular or experimental data: prepare features, train, tune, validate and interpret models — suited to predictive modelling in research.
Decide the dependent variable to predict, the available predictors, the sample unit and evaluation metrics.
A machine learning model should serve the research question first, rather than chase complex algorithms.
Handle missing values, outliers, categorical and continuous variables to produce a model-ready data table.
Training and validation data must go through the same feature processing pipeline.
Split into training, validation and test sets, or use cross-validation to assess model stability.
With small samples, cross-validation is usually more robust than a single split.
Train models such as random forest, gradient boosting and support vector machines, and do basic hyperparameter tuning.
Build a simple baseline first, then try more complex models.
Report metrics such as R², RMSE, MAE, accuracy, F1 or AUC, and analyse variable importance.
Describe how the model was validated in the paper; don't show training-set performance alone.
