Locomotor behaviour of Moina macrocopa exposed to permethrin: Potential for AI-based biological early warning systems

Abstract

Permethrin, a widely used pyrethroid pesticide, has emerged as a significant contaminant in aquatic ecosystems, with particularly high concentrations in Vietnam’s urban waterways. This study aims to develop a novel early warning system by assessing the sublethal behavioural effects of permethrin on the cladoceran Moina macrocopa and integrating these responses into an artificial neural network (ANN) model. A total of 21 individuals of M. macrocopawere exposed to seven permethrin concentrations (0–2.5 µg/L). Their swimming behaviour was video-recorded and analysed by using the motion-tracking software to extract a suite of locomotor parameters, which were then used to train a multi-layer ANN model. The results reveal a distinct biphasic (hormetic) dose-response: low concentrations (≤0.5 µg/L) stimulated swimming activity, whereas higher concentrations caused inhibition. In contrast, parameters such as turning angle and sinuosity increased proportionally with permethrin levels. The developed ANN model demonstrates high efficacy, achieving an overall accuracy of 94.05% on an independent test set for classifying four pollution levels. Notably, the model perfectly distinguished between non-polluted and highly polluted samples, confirming its reliability for detecting high-risk events. This study successfully establishes that integrating sublethal behavioural responses with artificial intelligence provides a sensitive, rapid, and automated tool for environmental monitoring. This approach holds significant potential for proactive water quality protection, especially in regions such as Vietnam, where automated biological monitoring systems are still underdeveloped. This system, enabling early detection of ecological stress through behavioural cues, can support local authorities in making timely, evidence-based decisions to mitigate pollution impacts and protect aquatic biodiversity.