CRAFTIIF: Cross-Resolution Analytic Four-Type Interpretable Isolation Forest for Multivariate Time Series Anomaly Detection

2026-06-11 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Mathematics & Computational Sciences · Depth: Expert, quick

Summary

CRAFTIIF (Cross-Resolution Analytic Four-Type Interpretable Isolation Forest) is a fully unsupervised framework designed for multivariate time series anomaly detection, specifically addressing four distinct anomaly types: point, distributional, temporal, and collective. It operates without dataset-specific tuning, generating K=500 random analytic wavelet features across four families (Morlet, DOG, Haar, Coiflet), each targeting a specific anomaly type. These features feed five structured Isolation Forests—one per type plus a meta-IF for compound anomalies—with an adaptive Otsu/MAD threshold calibrating detection for anomaly rates from 0.1% to 69.2%. CRAFTIIF provides direct anomaly-type attribution by construction. Evaluated on all 19 datasets of the mTSBench benchmark, it achieved a mean F1=0.228 (all 19 datasets) and F1=0.322 (13 detectable datasets), ranking first among 25 methods on VUS-PR (0.463 vs. previous best 0.329, a +40.7% improvement).

Key takeaway

For Machine Learning Engineers and Data Scientists tasked with multivariate time series anomaly detection, CRAFTIIF presents a significant advancement. You should consider this framework for its ability to detect all four distinct anomaly types with high accuracy and provide direct interpretability, outperforming 25 other methods on the mTSBench benchmark. Its unsupervised nature and lack of dataset-specific tuning make it a robust choice for diverse applications.

Key insights

CRAFTIIF addresses four distinct anomaly types in multivariate time series using specialized wavelet features and Isolation Forests for direct attribution.

Principles

• Anomaly types require distinct feature representations.
• Type-specific features enable direct anomaly attribution.
• Adaptive thresholding significantly improves detection.

Method

CRAFTIIF generates K=500 random analytic wavelet features across four families, feeding five structured Isolation Forests (one per type, one meta-IF), with adaptive Otsu/MAD thresholding.

In practice

• Use Morlet, DOG, Haar, Coiflet wavelets for specific anomaly types.
• Implement a meta-IF for compound anomaly detection.
• Apply adaptive Otsu/MAD for automatic threshold calibration.

Topics

• Multivariate Time Series
• Anomaly Detection
• Isolation Forest
• Wavelet Features
• Interpretability
• mTSBench Benchmark

Code references

• smitswil/craftiif

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Data Scientist

Related on AIssential

• CRAFTIIF: Cross-Resolution Analytic Four-Type Interpretable Isolation Forest for Multivariate Time Series Anomaly Detection — CRAFTIIF offers interpretable, unsupervised MTSAD by structurally separating anomaly types using wavelet features and dedicated Isolation Forests.
• Distribution-Free Stochastic Analysis and Robust Multilevel Vector Field Anomaly Detection — A distribution-free stochastic functional analysis method detects anomalies in vector fields using covariance and multilevel orthogonal subspaces.
• Data Machina #261 — Generative AI and LLMs are being adapted to achieve state-of-the-art performance in diverse time-series tasks.
• AnchorMoE: Interpretable Time Series Classification via Anchor-Routed MoE — AnchorMoE provides ante-hoc interpretable time series classification using an MoE architecture with specialized experts and noise suppression.
• IBAD: Interpretable Behavioral Anomaly Detection on Human Mobility Data — IBAD learns transferable, interpretable daily behavioral templates from human mobility data using LDA and self-supervised modeling.
• TRACE: A Temporal Conditional Estimation for Multimodal Time Series Foundation Models — TRACE leverages conditional diffusion for robust multimodal time series imputation, outperforming naive methods under missingness.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.