An evaluation of machine learning classifiers for next-generation, continuous-ethogram smart trackers

Hui Yu,Max Kröschel,Guozheng Li,Ran Nathan,Jian Deng,Marcel Klaassen,Sasha Pekarsky

doi:10.1186/s40462-021-00245-x

Abstract

BackgroundOur understanding of movement patterns and behaviours of wildlife has advanced greatly through the use of improved tracking technologies, including application of accelerometry (ACC) across a wide range of taxa. However, most ACC studies either use intermittent sampling that hinders continuity or continuous data logging relying on tracker retrieval for data downloading which is not applicable for long term study. To allow long-term, fine-scale behavioural research, we evaluated a range of machine learning methods for their suitability for continuous on-board classification of ACC data into behaviour categories prior to data transmission.MethodsWe tested six supervised machine learning methods, including linear discriminant analysis (LDA), decision tree (DT), support vector machine (SVM), artificial neural network (ANN), random forest (RF) and extreme gradient boosting (XGBoost) to classify behaviour using ACC data from three bird species (white stork Ciconia ciconia, griffon vulture Gyps fulvus and common crane Grus grus) and two mammals (dairy cow Bos taurus and roe deer Capreolus capreolus).ResultsUsing a range of quality criteria, SVM, ANN, RF and XGBoost performed well in determining behaviour from ACC data and their good performance appeared little affected when greatly reducing the number of input features for model training. On-board runtime and storage-requirement tests showed that notably ANN, RF and XGBoost would make suitable on-board classifiers.ConclusionsOur identification of using feature reduction in combination with ANN, RF and XGBoost as suitable methods for on-board behavioural classification of continuous ACC data has considerable potential to benefit movement ecology and behavioural research, wildlife conservation and livestock husbandry.

Highlights

Our understanding of movement patterns and behaviours of wildlife has advanced greatly through the use of improved tracking technologies, including application of accelerometry (ACC) across a wide range of taxa
Five of the six models (i.e. decision tree (DT), support vector machine (SVM), random forest (RF), artificial neural network (ANN) and Extreme gradient boosting (XGBoost)) generally had slightly lower accuracy when using a simplified compared to a full feature set, with a ~ 3.7% max mean accuracy difference
The relatively low variation in the F1 scores of the different classification methods within a certain behaviour in comparison to the variation across the different behaviours was striking, either with full feature set or simplified feature set (Fig. 2). This suggests that, some algorithms were clearly better than others, all machine learning methods had similar classification/ mis-classification issues

Summary

Introduction

Our understanding of movement patterns and behaviours of wildlife has advanced greatly through the use of improved tracking technologies, including application of accelerometry (ACC) across a wide range of taxa. In addition to the position of tracked animals in time, advanced biologging technologies provide opportunities for additional environmental data collection such as ambient temperature, light intensity and water depth, and data related to logger carriers such as heart rate, energy expenditure and behaviour [2,3,4]. Dynamic acceleration is due to changes of velocity caused by animal movement [8]. Based on these characteristics, at least four types of studies have been routinely conducted using ACC. Because animal behaviours consist of different postures and dynamic movement traits, ACC data has been used to classify animal behaviours (e.g. [7, 17,18,19])

Methods

Results

Discussion

Conclusion