EVEREST: An Evidential, Tail-Aware Transformer for Rare-Event Time-Series Forecasting

TL;DR

EVEREST is a transformer-based model designed for rare-event time-series forecasting, providing calibrated probabilistic predictions, tail risk estimation, and interpretability, suitable for high-stakes applications like space-weather and industrial monitoring.

Contribution

It introduces a novel, integrated transformer architecture with uncertainty and tail risk modeling, optimized for rare-event forecasting with high accuracy and efficiency.

Findings

Achieves state-of-the-art TSS of over 0.97 for space-weather flare prediction.

Efficiently trains with approximately 0.81 million parameters on commodity hardware.

Demonstrates applicability to high-stakes domains like weather and satellite diagnostics.

Abstract

Forecasting rare events in multivariate time-series data is challenging due to severe class imbalance, long-range dependencies, and distributional uncertainty. We introduce EVEREST, a transformer-based architecture for probabilistic rare-event forecasting that delivers calibrated predictions and tail-aware risk estimation, with auxiliary interpretability via attention-based signal attribution. EVEREST integrates four components: (i) a learnable attention bottleneck for soft aggregation of temporal dynamics; (ii) an evidential head for estimating aleatoric and epistemic uncertainty via a Normal--Inverse--Gamma distribution; (iii) an extreme-value head that models tail risk using a Generalized Pareto Distribution; and (iv) a lightweight precursor head for early-event detection. These modules are jointly optimized with a composite loss (focal loss, evidential NLL, and a tail-sensitive EVT penalty) and act only at training time; deployment uses a single classification head with no inference overhead (approximately 0.81M parameters). On a decade of space-weather data, EVEREST achieves state-of-the-art True Skill Statistic (TSS) of 0.973/0.970/0.966 at 24/48/72-hour horizons for C-class flares. The model is compact, efficient to train on commodity hardware, and applicable to high-stakes domains such as industrial monitoring, weather, and satellite diagnostics. Limitations include reliance on fixed-length inputs and exclusion of image-based modalities, motivating future extensions to streaming and multimodal forecasting.