Icon

Resampling in Supervised Fraud Detection Models

There has been no title set for this workflow's metadata.

This workflow shows the performances of four different decision tree models for fraud detection, each one trained on a different training set. One training set represents the original data with highly unbalanced target classes. The three others have been resampled to make the target classes balanced. The resampling methods that we demonstrate in this workflow are SMOTE oversampling, Bootstrap oversampling, and Bootstrap undersampling.

URL: Scoring Metrics eBook https://www.knime.com/knimepress/scoring-metrics-evaluating-machine-learning-models
URL: Resampling Imbalanced Data and Its Limits https://www.kdnuggets.com/2020/12/resampling-imbalanced-data-limits.html

Monitoring the performance of a decision tree model for fraud detection after resampling

Data access and preparation

Download the full dataset on Kaggle

Baseline model

Oversampling (SMOTE)

Oversampling (Bootstrap)

Undersampling (Bootstrap)

Decision Tree Predictor
Scorer (JavaScript)
Scorer (JavaScript)
Model 1: Baseline
Decision Tree Learner
undersampling legitimate
Bootstrap Sampling
Decision Tree Predictor
by target
Nominal Value Row Splitter
Scorer (JavaScript)
Model 4: Undersampling
Decision Tree Learner
Concatenate
oversampling fraudulent
Bootstrap Sampling
Decision Tree Predictor
Read credit card data (sample)
CSV Reader
Concatenate
by target
Nominal Value Row Splitter
Decision Tree Predictor
Model 3: Oversampling
Decision Tree Learner
80/20
Partitioning
Target column fraudulent/legitimate
Rule Engine
Scorer (JavaScript)
SMOTE
Model 2: SMOTE
Decision Tree Learner

Nodes

Extensions

Links