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

2026-06-12 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, extended

Summary

CRAFTIIF (Cross-Resolution Analytic Four-Type Interpretable Isolation Forest) is a novel, fully unsupervised framework for multivariate time series anomaly detection (MTSAD). It simultaneously addresses four distinct anomaly types: point, distributional, temporal, and collective, without requiring dataset-specific tuning. The system generates K=500 random analytic wavelet features from Morlet, DOG, Haar, and Coiflet families, feeding them into five structured Isolation Forests—one for each anomaly type and a meta-IF for compound anomalies. An adaptive hybrid threshold, combining Otsu bimodality and MAD estimation, automatically calibrates detection. Evaluated on all 19 mTSBench datasets, CRAFTIIF achieved a mean F1 of 0.228 overall and 0.322 on 13 detectable datasets, alongside a mean VUS-PR of 0.463, surpassing 24 baselines, including PCA (0.329) and raw IsolationForest (0.300). A diagnostic framework identifies 6 mTSBench datasets as fundamentally undetectable by unsupervised signal-statistics methods.

Key takeaway

For Machine Learning Engineers developing MTSAD systems, CRAFTIIF offers a robust, interpretable solution. You should consider its structured five-branch Isolation Forest architecture and adaptive thresholding to handle diverse anomaly types and rates without dataset-specific tuning. This approach provides direct anomaly-type attribution, crucial for root cause analysis and building trust in safety-critical deployments. Implement its diagnostic framework to identify fundamentally undetectable datasets, preventing wasted effort on intractable problems.

Key insights

CRAFTIIF offers interpretable, unsupervised MTSAD by structurally separating anomaly types using wavelet features and dedicated Isolation Forests.

Principles

• Anomaly type heterogeneity requires distinct feature representations.
• Direct anomaly-type attribution needs partitioned detection models.
• Adaptive thresholding significantly improves F1 across varied anomaly rates.

Method

CRAFTIIF extracts K=500 random analytic wavelet features per family (Morlet, DOG, Haar, Coiflet), feeds them to five type-specific Isolation Forests, and applies an adaptive Otsu/MAD hybrid threshold for detection.

In practice

• Use type-specific Isolation Forests for interpretable anomaly attribution.
• Implement adaptive thresholding for robust detection across anomaly rates.
• Employ diagnostic metrics to assess dataset detectability limits.

Topics

• Multivariate Time Series Anomaly Detection
• Isolation Forest
• Wavelet Features
• Anomaly Type Attribution
• Adaptive Thresholding
• 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 addresses four distinct anomaly types in multivariate time series using specialized wavelet features and Isolation Forests for direct attribution.
• The Data Points That Don’t Belong -And Why They Matter More Than You Think — Modern outlier detection leverages diverse techniques to find anomalies in complex, high-dimensional, and streaming datasets.
• RFX-Fuse: Breiman and Cutler's Unified ML Engine + Native Explainable Similarity — RFX-Fuse unifies diverse ML tasks into a single Random Forest model, providing native explainability and GPU acceleration.
• KRONE: Scalable LLM-Augmented Log Anomaly Detection via Hierarchical Abstraction — Krone leverages hierarchical log structures and hybrid LLM-pattern matching for efficient, accurate, and interpretable anomaly detection.
• ProtoX-AD: Self-Explainable Time Series Anomaly Detection and Characterization — ProtoX-AD integrates interpretable prototypes into self-supervised time series anomaly detection for explainable results and anomaly characterization.
• Masked Diffusion Modeling for Anomaly Detection — MaskDiff-AD uses masked diffusion models to detect anomalies in discrete data by scoring reconstruction difficulty without reverse-time sampling.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.