Unsupervised-explainable anomaly detection in large-scale estuarine acoustic telemetry data

Abstract

Acoustic telemetry data play a vital role in understanding the behaviour and movement of aquatic animals. However, these datasets, which often consist of millions of individual detections, frequently contain anomalous movements that pose significant challenges. Traditionally, anomalous movements can be identified either manually or through basic statistical methods, approaches that are time-consuming and prone to high rates of unidentified anomalies in large datasets. This study focuses on the development of automated classifiers for a large telemetry dataset comprising detections from 50 acoustically tagged dusky kob ( Argyrosomus japonicus ) monitored in the Breede Estuary, South Africa. Using an array of 16 acoustic receivers deployed throughout the estuary between 2016 and 2021, resulting in the collection of over three million individual detections. This paper presents detailed guidelines for data pre-processing, resampling strategies, labelling process, feature engineering, data splitting methodologies, and the selection and interpretation of machine learning (ML) and deep learning (DL) models for anomaly detection. Among the evaluated models, neural network autoencoder (NN-AE) demonstrated superior performance, aided by our proposed threshold-finding algorithm. NN-AE achieved a high recall with no false normals (i.e., no misclassifications of anomalous movements as normal patterns), a critical factor in ensuring that no true anomalies are overlooked. In contrast, other models exhibited false normal fractions exceeding 0.9, indicating they failed to detect the majority of true anomalies—a significant limitation for telemetry studies where undetected anomalies can distort interpretations of movement patterns. While the NN-AE’s performance highlights its reliability and robustness in detecting anomalies, it faced challenges in accurately learning normal movement patterns when these patterns gradually deviated from anomalous ones. To the best of our knowledge, this study represents the first effort to develop automated methods leveraging ML and DL to address anomalous detections in acoustic telemetry data. • Introduced an automated threshold-finding algorithm for neural network autoencoders (NN-AE) that maximizes recall while minimizing false anomalies. • NN-AE outperformed traditional models (Isolation Forest, DBSCAN, LOF) with zero false normal, and negligible false anomalies, critical for ecological data integrity. • Proposed a Shannon frequency-based resampling method to handle irregular telemetry sampling, improving model stability and performance. • Engineered features (e.g., duration at stations, consecutive missed detections) aligned with expert-defined anomaly criteria (e.g., mortality signals, tag malfunctions). • Demonstrated efficacy on 3 million detections from 50 dusky kob, offering a scalable solution for big-data telemetry studies. • Detected anomalies linked to ecological events (e.g., predation, habitat degradation), aiding real-time ecosystem monitoring and management.