Time-Aware Validation of Machine Learning Ship Fuel Consumption Models (arXiv 2608.16833)

Forum topic · 小凯 · 2026-08-19

Summary

This paper addresses a critical flaw in machine learning models for ship fuel consumption (SFC) prediction: most prior studies use random train-test splits on high-frequency sensor data, which introduces temporal leakage and produces overly optimistic performance estimates. The authors apply time-aware evaluation methods—Time Series Cross-Validation (TSCV) and Blocked TSCV (BTSCV)—using the Canadian Coast Guard Ship Sir Wilfrid Laurier as a case study. Six regression models plus a physics baseline were tuned under three time-aware schemes and three feature configurations, then evaluated on a common temporally ordered hold-out set drawn from approximately 3.88 million steady-state 1Hz records. The work supports vessel operation optimization, emissions estimation, and decision support systems for sustainable maritime transportation. It highlights that realistic deployment conditions require chronological validation rather than random splits. Published on arXiv as 2608.16833 (2026-08-17) by Samarasimha Reddy Chittamuru, Ayhan Akinturk, Allison Kennedy, et al.

Overview

This arXiv paper (2608.16833), published 2026-08-17 by Samarasimha Reddy Chittamuru, Ayhan Akinturk, Allison Kennedy, et al., examines a common validation flaw in data-driven ship fuel consumption (SFC) models.

Key Points

Ship fuel consumption (SFC) prediction supports vessel operation optimisation, emissions estimation, and decision support systems (DSS) for sustainable maritime transportation.
Numerous data-driven fuel models have been developed over the past two decades, but most studies evaluate performance using random train-test splits.
Applied to high-frequency records, random splits admit temporal leakage and yield optimistic results that do not reflect deployment conditions.
The authors address this gap using time-aware evaluation: Time Series Cross-Validation (TSCV) and Blocked TSCV (BTSCV).
Case study: Canadian Coast Guard Ship (CCGS) Sir Wilfrid Laurier.
Six regression models and a physics baseline were tuned under three time-aware schemes and three feature configurations, then evaluated on a common temporally ordered hold-out set.
The evaluation set was extracted from approximately 3.88 million steady-state 1Hz records.

Original Abstract (excerpt)

> Ship fuel consumption (SFC) prediction supports vessel operation optimisation, emissions estimation, and decision support systems (DSS) for sustainable maritime transportation. Numerous data-driven fuel models have been developed over the past two decades, but a critical and often overlooked limitation lies in their validation practices: most studies evaluate performance using random train-test splits, which, applied to high-frequency records, admit temporal leakage and yield optimistic results that do not reflect deployment conditions. This paper examines that gap using time-aware evaluation, specifically Time Series Cross-Validation (TSCV) and Blocked TSCV (BTSCV). Using the Canadian Coast Guard Ship (CCGS) Sir Wilfrid Laurier as a case study, six regression models and a physics baseline...

Paper link: https://arxiv.org/abs/2608.16833