Advancing forest fires classification using Neurochaos learning

Pant, Kunal Kumar and Ajai AS, Remya and Nagaraj, Nithin (2026) Advancing forest fires classification using Neurochaos learning. IEEE Access, 14. pp. 145861-145892. ISSN 2169-3536

Text
Advancing_Forest_Fires_Classification_Using_Neurochaos_Learning.pdf - Published Version
Download (11MB)

Contribution	Name	Email

Abstract:	Forest fires are among the most dangerous and unpredictable natural disasters worldwide, with severe environmental and economic impacts motivating extensive research on fire-occurrence prediction using environmental variables. Although Machine Learning (ML) and Deep Learning (DL) methods are widely used for this task, their performance can degrade under limited training data, motivating nonlinear feature-transformation approaches for small and heterogeneous datasets. This work evaluates ChaosNet and Random Heterogeneous Neurochaos Learning (RHNL), two variants of Neurochaos Learning (NL), a chaos-based, brain-inspired learning paradigm previously shown to capture nonlinear patterns from limited data and preserve causal feature behaviour under controlled settings. Experiments are conducted across the Algerian Forest Fires (AFF), Canadian Forest Fires (CFF), and Portugal Forest Fires (PFF) datasets under high-sample and low-sample regimes, using random, time-aware, and spatially grouped validation where appropriate. On AFF, RHNL50G50L+SVM achieves a time-aware high-sample macro-F1 of 0.995±0.010 , while ChaosNet attains 0.816±0.04 macro-F1 using only 1% training data under random evaluation. On CFF, Random Forest is strongest in the high-sample setting ( 0.740±0.024 ), whereas ChaosNet leads at the 1% fraction in the log-transformed setting ( 0.568±0.03 ). On PFF, spatially grouped evaluation reduces performance, highlighting the importance of leakage-resistant validation for geographic generalization. Across datasets, ChaosNet trains in only 0.151– 0.205(ms) at the 1% fraction, making it approximately 360– 910× faster than Random Forest and 700– 1250× faster than Bagging DT.
Item Type:	Journal Paper
Subjects:	School of Natural and Engineering Sciences > Complex Systems
Divisions:	Schools > Natural Sciences and Engineering
Date Deposited:	30 Sep 2026 08:57
Last Modified:	01 Oct 2026 08:41
Official URL:	https://ieeexplore.ieee.org/document/11692837
Related URLs:
Funders:	*
Projects:	*
DOI:	https://doi.org/10.1109/ACCESS.2026.3734132
URI:	http://eprints.nias.res.in/id/eprint/3468

Actions (login required)

View Item