Icon

Big Data Fraud Detection Pipeline

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

This workflow streamlines fraud detection by integrating data importation, preprocessing, and sophisticated feature engineering with machine learning. It focuses on analyzing transaction data, highlighting anomalies, and calculating key metrics for fraud risk assessment. Emphasizing customer behavior and terminal risk over varied time frames, it leads to the development of an accurate, reusable machine learning model. This No-Code/Low-Code pipeline adeptly navigates the intricacies of large-scale fraud analysis.

URL: Reproducible Machine Learning for Credit Card Fraud Detection - Practical Handbook https://fraud-detection-handbook.github.io/fraud-detection-handbook/Foreword.html

Machine Learning Training Pipeline

B - Data Cleaning & Feature Engineering - Creation of new variables
C - Fraud Prediction and Model Saving on the Server

Big Data Pipeline Implementation - Case Study of Fraud Prediction with KNIME



The workflow we present today aims to address this issue by using a series of calculated steps to transform raw data into a fraud prediction model in a No-Code/Low-Code environment.

The key steps are:

  • Data Reading: Importing raw data.

  • Preprocessing & Feature Engineering: Transforming date and time fields and creating relevant indicators.

  • Predictions: Using a machine learning model and then saving the model to be able to reuse it in a second workflow.

Each node has its own function in the overall execution. More than 4000 nodes exist in the ecosystem, for all use cases.

Source: Treatments adapted from work in Python: By Machine Learning Group (Free University of Brussels - ULB).


A - Data Reading Node
Reading Random DataCSV of transactions
CSV Reader
Extraction of the dayand time of transactions
Extract Date&Time Fields
Date conversionstring to DateTime format
String to Date&Time
Flag transactionsduring the weekend
Rule Engine
Flag transactionsat night (midnight to 6am)
Rule Engine
Reading terminals
Excel Reader
Two functions for adding features:1 - Nb of transactions and Avg spending2 - Fraud risk for each terminalExtraction of complex relationshipsMetanode for reusability & sharing
Transformations Clients & Terminaux
Saving the modelfor reuse without retrainingon all data
PMML Writer
Model conversionto PMML format
Gradient Boosted Trees to PMML
Handles missing values
Missing Value
Dataset Split80%/20% Train/TestHold-Out Approach
Partitioning
Visualize fraud points
OSM Map View
Sampling to manageclass imbalanceEnsuring representativeness
Equal Size Sampling
Fraud ratioper dispenser
Geo View
Model trainingMachine Learning
Gradient Boosted Trees Learner
Predictionson the test setfrom the trained model
Gradient Boosted Trees Predictor
Reading the matrix to evaluatethe effectiveness of the modelAccuracy = 78%
Scorer (JavaScript)

Nodes

Extensions

Links