Anomaly detection unsupervised
Approcci unsupervised, distance-based, density-based, isolation forest.
Clustering: KMeans & GMM
Differenze KMeans vs GMM, quando usarli per anomaly detection.
Feature engineering temporale
Aggregati su finestre, rolling stats, lag features.
Soglia & metriche
Percentile-based threshold, F1, ROC-AUC, PR-AUC su classi sbilanciate.
Split temporale & no leakage
Time-aware split, prevenzione del leakage in time series.