Anomaly Detection using Autoencoders
summary
Autoencoder pipeline uses Spark to H2O conversion, trains an h2o deeplearning estimator autoencoder on the open credit‑card dataset (0.17% fraud), saves the model and scores reconstruction error per feature. A Qliksense dashboard highlights anomalies with a chosen threshold 0.01, flagging 105 anomalous transactions; 69 of 89 labeled fraud cases are detected. The workflow surfaces per‑feature reconstruction errors for investigation and is presented as a deployable, interpretable prototype that balances detection and business capacity.
AI-written summary · about
chapters (11)
- 0:00Intro
- 1:04TABLE OF CONTENTS
- 2:37INTRODUCTION TO AUTOENCODERS
- 4:32Uses of Autoencoders
- 5:47Restricted Boltzmann Machine
- 9:1703 OVER COMPLETE TECHNIQUES
- 10:17PYTHON LIBRARIES
- 11:10Examples of Fraudulent Acts
- 11:54Japan ATM Scam
- 13:50Anomaly vs Fraud
- 22:15Key Takeaways from TELCO
related talks
18mReal-time Fraud Detection using Deep Learning
19mThe Rise of AI in Anomaly Detection
45mAnomaly detection with TensorFlow
40mFraud detection without labeled data
25mUnsupervised shallow learning for fraud detection on marketplaces
29mAI-Powered Anomaly Detection in Cloud Data Pipelines
01h 18mIntroduction to H20 AutoML with Python
52mModern Fraud Prevention Using Deep Learning