DataHour: Model Explainability and Model Diagnosis

Model explainability is very important for machine learning models for regulatory compliance, ethical considerations and risk management. This session covers the importance of explaining your model and briefly discusses the following techniques:

1. Shapley Values 
2. Local Surrogate 
3. PDP/ICE Plots 
4. Surrogate Models
5. Counterfactuals

While explaining each of these techniques with examples, the presentation also compares these techniques. We will also see how we can use model explainability to diagnose and debug our models.

Srimugunthan is currently working as Data science-Lead in WellsFargo. Srimugunthan is a data analytics professional with 12+ years experience in tech and 8+ years experience in Big data analytics. His expertise is in Prediction models, Recommender systems, Reinforcement learning, Experimentation and Causal Inference. He has worked in Finance and consumer internet domains and enjoys solving Data Science problems and finding data science use cases in these domains. He has completed his masters in computer science from Indian Institute of science and has an Executive MBA in product management.