JANET: Joint Adaptive predictioN-region Estimation for Time-series

2024-07-08

arXiv.org

AI Summary

Existing conformal prediction methods rely on the exchangeability assumption, which is fundamentally incompatible with uncertainty quantification in multi-step time series forecasting. This work proposes the first joint conformal prediction framework specifically designed for multivariate time series, circumventing the exchangeability requirement and constructing statistically valid joint prediction regions. Methodologically, it integrates generalized inductive conformal prediction, rolling calibration, sequence-dependent modeling, and joint confidence set optimization, and introduces a novel K-family error rate control mechanism enabling task-driven, adaptive region adjustment. Evaluated on multiple real-world time series benchmarks, the approach significantly improves both coverage accuracy and interval sharpness for multi-step forecasts—achieving strong theoretical guarantees without compromising practical utility.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Relational Probabilistic Models
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Security and Privacy: Large-scale security measurements
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

Conformal prediction provides machine learning models with prediction sets that offer theoretical guarantees, but the underlying assumption of exchangeability limits its applicability to time series data. Furthermore, existing approaches struggle to handle multi-step ahead prediction tasks, where uncertainty estimates across multiple future time points are crucial. We propose JANET (Joint Adaptive predictioN-region Estimation for Time-series), a novel framework for constructing conformal prediction regions that are valid for both univariate and multivariate time series. JANET generalises the inductive conformal framework and efficiently produces joint prediction regions with controlled K-familywise error rates, enabling flexible adaptation to specific application needs. Our empirical evaluation demonstrates JANET's superior performance in multi-step prediction tasks across diverse time series datasets, highlighting its potential for reliable and interpretable uncertainty quantification in sequential data.

Problem

• Extends conformal prediction to time series data
• Handles multi-step ahead prediction tasks
• Provides joint prediction regions for uncertainty quantification

Innovation

• Adaptive conformal prediction for time series
• Joint multi-step prediction regions
• Controlled K-familywise error rates