Considerations for selecting decision trees for binary classification problems.
The shippingData_decisionTree.ipynb notebook shows how to perform model selection for decision trees in binary classification problems. It is divided in the following sections:
- A. Decision Tree (DT): Summary
- B. Data inspection & preparation
- C. Model selection
- D. Model predictions & interpretation
It uses as an example an e-commerce dataset composed of 10999 purcheses (rows), 11 features, and a binary target (on time / not on time). Target represents whether a product has been delivered on time (1) or not (0). Features include information regarding customers, product, and shipment. The dataset can be downloaded from Kaggle: https://www.kaggle.com/datasets/prachi13/customer-analytics
The notebook focuses on deciding on whether to (i) over-/under-sample the data in case of imbalanced datasets, (ii) use gini or entropy as criteria, and (iii) limiting tree depth.
