Why Your AI Model Lies: Debugging Healthcare Predictions with SHAP and Python

Machine learning models are increasingly used in healthcare, but what happens when they fail silently?

In this talk, we walk through a real Python-based cardiovascular disease (CVD) risk prediction system and uncover how models can produce misleading and counterintuitive results when faced with unexpected or invalid inputs. Using tools like scikit-learn and SHAP, we demonstrate how to interpret model predictions, identify hidden issues, and understand why explanations are not always straightforward—especially in multi-class settings.

You’ll learn how to:

• Build and interpret ML models in Python
• Use SHAP to explain predictions
• Detect unreliable or out-of-distribution inputs
• Improve trust in real-world AI systems

Includes a live demo of a real-world prediction system.

This session is beginner-friendly and ideal for Python developers interested in machine learning, data science, or building reliable AI systems.

Presented by

Krishna Mridha