Robust State-Conditional Feature-Weighted Jump Models for Temporal Clustering

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

Summary

A Robust State-Conditional Feature-Weighted Jump Model (FWJM) is introduced for time-dependent clustering, incorporating a Tukey's biweight loss function for robustness against outliers and a penalty to encourage smooth temporal transitions. This model also features an additional parameter that controls the variability of feature weights across different states, allowing for state-specific feature relevance. Simulation studies, including scenarios with T=1000 observations and P=5 or P=50 features, demonstrate that FWJM accurately recovers true cluster sequences and reliably identifies relevant features, consistently outperforming competing methods, especially in the presence of 5% data contamination. Empirical applications include analyzing daily conflict-related homicides in Kosovo from 1998-2000 (T=1081, P=3) and macroeconomic performance indicators for twelve European countries from 1949-2024 (T=49, P=36).

Key takeaway

For Machine Learning Engineers analyzing multivariate time series with dynamic regimes and potential outliers, you should consider implementing Robust Feature-Weighted Jump Models. This approach provides superior clustering accuracy and identifies state-specific feature relevance, which is crucial for understanding underlying dynamics. Tune hyperparameters like temporal persistence (λ) and feature weight entropy (ζ) based on your data's time series length and feature sparsity to optimize performance.

Key insights

Robust Feature-Weighted Jump Models accurately cluster time series and identify state-specific feature importance, even with outliers.

Principles

• State-specific feature weights enhance temporal clustering.
• Robust loss functions handle time series outliers effectively.
• Temporal regularization improves latent state sequence persistence.

Method

FWJM minimizes an objective function combining dissimilarity to medoids, an entropy penalty for weights, and a temporal persistence penalty. Estimation uses an alternating optimization procedure with three steps: updating medoids, state sequence via dynamic programming, and weights in closed form.

In practice

• Use Tukey's biweight loss for outlier robustness.
• Select hyperparameters via Silhouette index.
• Employ multi-start optimization for global optimum.

Topics

• Temporal Clustering
• Feature Weighting
• Robust Statistics
• Time Series Analysis
• Jump Models
• Multivariate Data

Code references

• FedericoCortese/FWJM

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

Related on AIssential

• Generative Modeling of Approximately Periodic Time Series by a Posterior-Weighted Gaussian Process — A two-stage GP model generates stable, approximately periodic time series by decoupling intra- and inter-repetition variability.
• Learning with Shallow Neural Networks on Cluster-Structured Features — Input feature correlations significantly reduce sample complexity for shallow neural networks under specific conditions.
• Estimating Mutual Information between Time Series and Temporal Event Sequences Across Diverse Analysis Tasks — A nonparametric mutual information estimator quantifies dependence between heterogeneous time series and event sequences without transformation or discretization.
• Scalable Model-Based Clustering with Sequential Monte Carlo — A novel SMC algorithm improves scalability for online clustering by decomposing problems into independent subproblems.
• Continual Fine-Tuning with Provably Accurate and Parameter-Free Task Retrieval — Proteus offers provably accurate, parameter-free task retrieval for continual fine-tuning by leveraging distinct representation signatures.
• The Feature Selection Trap: Why ‘More Data’ Can Actively Hurt Your Machine Learning Model — Adding more features does not always improve ML model performance and can actively degrade it due to redundancy.

Open in AIssential →

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