The present disclosure provides a novel system for multi-level multi-variate embedding framework for risk scoring. In the present disclosure, risk factors within each category are aggregated as one score using Level-1 embedding, while a global risk score is obtained using Level-2 embedding. The framework models different scores considering the interactive effects among different factors and scores. Also, any auxiliary variables that are important for global risk scoring can be easily integrated. The coefficients in multi-level embedding can be determined within one process based on historical data of user interested activities. A penalized loss is used to optimize the model structure and to avoid overfitting. Popular classification model evaluation metrics can be used to check the performance of embedding scores.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.