Compliance monitoring · Production

Netflix applies machine learning for real-time fraud detection in streaming services

The problem

Detecting content fraud, service fraud, and account fraud at scale in real time is highly challenging for streaming platforms. Rule-based detection approaches require constant expert supervision, cannot scale for real-time use, and may become biased or limited in scope.

Workflow diagram · grounded in source
1
License and manifest request
trigger
“a request is sent to the streaming server to obtain the protected encrypted digital content. In order to stream the digital content, the user requests a license from the clearinghouse that verifies the user's credentials”
2
Heuristic-based data labeling
validation
“we define a set of rule-based heuristics used for identifying anomalous streaming behaviors of clients and label them as anomalous or benign”
3
Streaming behavior feature extraction
ai_action
“we define features based on the expected streaming behavior of the users and their interactions with devices”
4
SMOTE class imbalance treatment
ai_action
“we use the Synthetic Minority Over-sampling Technique (SMOTE) to over-sample the minority classes by creating a set of synthetic samples”
5
ML anomaly detection inference
ai_action
“model-based and data-driven anomaly detection strategies to identify them”
6
Fraud classification output
output
“the deep auto-encoder model performs the best among the semi-supervised anomaly detection approaches with an accuracy of around 96% and f1 score of 94%”
Reported outcome

The deep auto-encoder model outperformed all other semi-supervised anomaly detection approaches, achieving around 96% accuracy and a 94% f1 score on streaming fraud classification.

Reported metrics
Deep auto-encoder accuracyaround 96%
Deep auto-encoder f1 score94%
Benign accounts in training dataset1,030,005
Anomalous accounts in training dataset28,045
Reported stack
deep auto-encoderSMOTEOne-Class SVMIsolation ForestXGBoostRandom ForestGradient BoostingDRM
Source
https://netflixtechblog.com/machine-learning-for-fraud-detection-in-streaming-services-b0b4ef3be3f6
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

The deep auto-encoder model outperformed all other semi-supervised anomaly detection approaches, achieving around 96% accuracy and a 94% f1 score on streaming fraud classification.

What tools did this team use?

deep auto-encoder, SMOTE, One-Class SVM, Isolation Forest, XGBoost, Random Forest, Gradient Boosting, DRM.

What results were reported?

Deep auto-encoder accuracy: around 96%; Deep auto-encoder f1 score: 94%; Benign accounts in training dataset: 1,030,005; Anomalous accounts in training dataset: 28,045 (source-reported, not independently verified).

How is this compliance monitoring AI workflow structured?

License and manifest request → Heuristic-based data labeling → Streaming behavior feature extraction → SMOTE class imbalance treatment → ML anomaly detection inference → Fraud classification output.

Related compliance monitoring cases