ML Anomaly Detection

Use this skill when detecting anomalies or outliers in data, building unsupervised anomaly detection systems, applying statistical methods (Z-score/IQR), proximity-based (LOF), ensemble (Isolation Forest), deep learning (autoencoder/VAE), or time-series anomaly detection. This skill enforces: method selection by data characteristics (tabular/time-series/high-dim), statistical baseline (Z-score/IQR), model configuration (contamination rate, threshold), evaluation with precision/recall at k, real-time pipeline design. Do NOT use for: supervised fraud detection with labeled data (use classification skill), data quality checks (use data-validation skill), root cause analysis …