Tuesday, 30 March 2027
12:45 | Join in |
13:00 | Welcome, Motivation & Introduction |
13:15 | Recap: Explainable AI (XAI) for anomaly detection and time-dependent data |
– global vs. local explanations | |
– model-agnostic vs. model-specific methods | |
– model justification | |
– specifics for time series (temporal dependence, feature engineering) | |
13:30 | Guided end-to-end XAI exercise on transaction data (Credit Card Fraud) |
– exploratory data analysis | |
– understanding data characteristics (imbalance, anonymized features) | |
– anomaly detection / classification modelling | |
– explanation methods (feature importance, SHAP) | |
Hands-on lab: training a lightweight model and interpreting fraud predictions | |
14:15 | Break |
14:30 | Guided end-to-end XAI exercise on time series data (NYC Taxi) |
– time series exploration (trend, seasonality, anomalies) | |
– modelling (baseline + anomaly detection via residuals / Orion reference) | |
– explanation methods for temporal data (lag features, residual analysis, SHAP) | |
Hands-on lab: detecting and explaining anomalies in time series | |
15:15 | Break |
15:45 | Comparison, limitations & discussion |
– transferability of XAI methods across data types | |
– limitations of explanations (data, features, models) | |
– interpreting anomalies vs. model behavior | |
– limitations in explainability | |
Interactive session: brainstorming and discussion on trust, usability, and open challenges | |
17:00 | End of course |