Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Multi-task CNN Behavioral Embedding Model For Transaction Fraud Detection

Bo Qu, Zhurong Wang, Minghao Gu, Daisuke Yagi, Yang Zhao, Yinan Shan, Frank Zahradnik

TL;DR

This paper tackles transaction fraud detection in e-commerce by framing it as multivariate time series analysis of user behavior and proposing a lightweight Multi-task CNN Behavioral Embedding Model (MTCNN). The approach uses a single-layer CNN with multi-range kernels, augmented by positional encoding and a scalable universal embedding for continuous and categorical features, trained with hard parameter sharing and Random Loss Weighting to balance multiple fraud-related tasks. Empirical results on real-world data show MTCNN is competitive with Transformer-based models like TST in KS/IV and offers robust downstream gains when its embeddings are used in GBM-based detectors, with a smaller parameter footprint. The work highlights practical benefits for near-real-time deployment and suggests further integration of CNN and multitask techniques to broaden capabilities in fraud detection.

Abstract

The burgeoning e-Commerce sector requires advanced solutions for the detection of transaction fraud. With an increasing risk of financial information theft and account takeovers, deep learning methods have become integral to the embedding of behavior sequence data in fraud detection. However, these methods often struggle to balance modeling capabilities and efficiency and incorporate domain knowledge. To address these issues, we introduce the multitask CNN behavioral Embedding Model for Transaction Fraud Detection. Our contributions include 1) introducing a single-layer CNN design featuring multirange kernels which outperform LSTM and Transformer models in terms of scalability and domain-focused inductive bias, and 2) the integration of positional encoding with CNN to introduce sequence-order signals enhancing overall performance, and 3) implementing multitask learning with randomly assigned label weights, thus removing the need for manual tuning. Testing on real-world data reveals our model's enhanced performance of downstream transaction models and comparable competitiveness with the Transformer Time Series (TST) model.

Multi-task CNN Behavioral Embedding Model For Transaction Fraud Detection

TL;DR

This paper tackles transaction fraud detection in e-commerce by framing it as multivariate time series analysis of user behavior and proposing a lightweight Multi-task CNN Behavioral Embedding Model (MTCNN). The approach uses a single-layer CNN with multi-range kernels, augmented by positional encoding and a scalable universal embedding for continuous and categorical features, trained with hard parameter sharing and Random Loss Weighting to balance multiple fraud-related tasks. Empirical results on real-world data show MTCNN is competitive with Transformer-based models like TST in KS/IV and offers robust downstream gains when its embeddings are used in GBM-based detectors, with a smaller parameter footprint. The work highlights practical benefits for near-real-time deployment and suggests further integration of CNN and multitask techniques to broaden capabilities in fraud detection.

Abstract

The burgeoning e-Commerce sector requires advanced solutions for the detection of transaction fraud. With an increasing risk of financial information theft and account takeovers, deep learning methods have become integral to the embedding of behavior sequence data in fraud detection. However, these methods often struggle to balance modeling capabilities and efficiency and incorporate domain knowledge. To address these issues, we introduce the multitask CNN behavioral Embedding Model for Transaction Fraud Detection. Our contributions include 1) introducing a single-layer CNN design featuring multirange kernels which outperform LSTM and Transformer models in terms of scalability and domain-focused inductive bias, and 2) the integration of positional encoding with CNN to introduce sequence-order signals enhancing overall performance, and 3) implementing multitask learning with randomly assigned label weights, thus removing the need for manual tuning. Testing on real-world data reveals our model's enhanced performance of downstream transaction models and comparable competitiveness with the Transformer Time Series (TST) model.

Paper Structure

This paper contains 17 sections, 9 equations, 2 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. MTS modeling
  4. Enhancing Generalizability with Multi-task Learning
  5. Approach
  6. MTS Data Notation and Embedding
  7. Positional Encoding
  8. Convolutional Neural Network
  9. Multi-task Learning
  10. Architecture Advantages
  11. Experiment
  12. Datasets and Evaluation Metrics
  13. Experimental Results and Analysis
  14. Stage 1
  15. Stage 2
  16. ...and 2 more sections

Figures (2)

  • Figure 1: MTCNN model architecture with three channels as example.
  • Figure 2: Illustration of user behavior flow of page browsing.