William Smits Β· Jun 11, 2026
CRAFTIIF (Cross-Resolution Analytic Four-Type Interpretable Isolation Forest) is a fully unsupervised framework that detects four types of anomalies in multivariate time series: point, distributional, temporal, and collective. CRAFTIIF achieves high performance on 19 datasets from the mTSBench benchmark, ranking first among evaluated methods, and offers direct anomaly-type attribution through its branch firing mechanism. The framework is adaptive, using an Otsu/MAD threshold that calibrates detection across a wide range of anomaly rates.
Why This Matters
This paper's approach to developing a cross-resolution anomaly detection framework can be directly applied to power system engineers to identify and mitigate unusual patterns in grid data, such as rogue states or unusual frequency fluctuations, which can impact grid stability and operation. By enhancing the ability to detect anomalies, CRAFTIIF has practical implications for grid operators and utility planners in optimizing grid performance and ensuring reliability.