Evaluating ROCKET and Catch22 features for calf behaviour classification from accelerometer data using Machine Learning models

Aug 4, 2026

Upload by Noah

Continuous monitoring of calf behaviour can support identification of routine practices that affect dairy-farm welfare. Accelerometer signals from neck collars can be paired with machine learning to automate behavioural classification, but robust generalization across animals is still challenging. This work compares ROCKET and Catch22 time-series feature sets against hand-crafted features. Using annotated accelerometer sequences from 30 pre-weaned calves, models classify six welfare-relevant behaviours and are evaluated for balanced accuracy on held-out data.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages45

Reading Time1 h 53 min

AccessFree to download

Abstract

Introduction

Calf welfare and stress factors

Need for automated behavioural monitoring

Accelerometer-based sensing

Machine learning feature approaches

arXiv :2404 . 18159v2 [ cs .LG] 30 Apr 2024 Evaluating ROCKET and Catch22 features for calf behaviour classification from accelerometer data using Machine Learning models Oshana Dissanayakea,c , Sarah E. McPhersonc,d,e , Joseph Allyndr´eeb , Emer Kennedyc,d , P´adraig Cunninghama , Lucile Riaboffa,c,f a School of Computer Science, University College Dublin, Ireland b School of Maths and Stats, University College Dublin, Ireland c VistaMilk SFI Research Centre, Ireland dTeagasc, Animal & Grassland Research and Innovation Centre, Moorepark, Fermoy, Co. Cork, P61C997, Ireland e Animal Production Systems Group, Wageningen University & Research, Wageningen, The Netherlands f GenPhySE, Universit´e de Toulouse, INRAE, ENVT, 31326, Castanet-Tolosan, France Abstract Monitoring calf behaviour continuously would be beneficial to identify routine practices (e.g., weaning, transport, dehorning, etc.) that impact calf welfare in dairy farms. In that regard, accelerometer data collected from neck collarscan be used along with Machine Learning models to classify calf behaviour automatically. However, further development is needed to classify a broad spectrum of behaviours with good genericity from one animal to another. While Hand-Crafted features are typically used in the field as inputs for Machine Learning models, feature sets designed explicitly for time-series classification problems have been developed in related fields, such as ROCKET and Catch22 features. This study aims to compare the performance of ROCKET and Catch22 features to Hand-Crafted features commonly used in the field. 30 Irish Holstein Friesian and Jersey pre-weaned calves were equipped with an accelerometer sensor for several weeks, and their behaviours were annotated, allowing for 27.4 hours of observation aligned with the accelerometer time-series. Additional time-series were computed from the raw X, Y and Z-axis and split into 3-second time windows. ROCKET, Catch22 and HandCrafted features were calculated for each time window, and the dataset was then split into the train, validation and test sets. Each set of features was used to train three Machine Learning models (Random Forest, eXtreme Gra- dient Boosting, and RidgeClassifierCV) to classify six behaviours indicative of pre-weaned calf welfare (drinking milk, grooming, lying, running, walking and other) . Models were tuned with the validation set, and the performance of each feature-model combination was evaluated with the test set. The best performance across the three models was obtained with ROCKET [average balanced accuracy ± standard deviation](0.70 ± 0.07), followed by Catch22 (0.69 ± 0.05), well ahead of Hand-Crafted (0.65 ± 0.034) . The best balanced accuracy (0.77) was obtained with ROCKET and RidgeClassifierCV, followed by Catch22 and Random Forest (0.73) . Thus, tailoring these approaches for specific behaviours and contexts will be crucial in advancing precision livestock farming and enhancing animal welfare on a larger scale. Keywords: Dairy calf, Behavior, Accelerometers, ROCKET, Catch22, Machine Learning, Features 1. Introduction Enhancing the welfare of young farm animals through the adoption of suitable practices is likely to improve their performance at different scales. Indeed, the prolonged effects of stress can lead to an exhaustion phase, resulting in decreased performance, an increased risk of disease, and slowed growth. In particular, calves are subjected to many stressful events from their first weeks (dehorning, weaning, transport, social isolation, relocation, etc.) . Improving calf welfare is thus highly important to prevent physiological changes and death that may happen due to prolonged exposure to stress (Koknaroglu and Akunal, 2013) but also to bridge the gap between farming and society, as consumers place animal welfare as one of their primary expectations (Cardoso et al., 2016) . Changes in calf behaviour, such as altered feeding patterns, decreased p

FAQs

What is the purpose of comparing ROCKET and Catch22 features with hand-crafted features?

The study evaluates whether ROCKET and Catch22 time-series feature sets can outperform hand-crafted features for classifying calf behaviours from accelerometer data and achieve better general performance across animals.

How was the accelerometer dataset collected and prepared?

Thirty pre-weaned Holstein Friesian and Jersey calves wore neck-collar accelerometers for several weeks, producing annotated behaviour observations. Raw X, Y and Z signals were transformed into time-series and split into 3-second windows aligned with the annotations.

Which model and feature combination achieved the best balanced accuracy?

ROCKET features produced the best overall performance across the tested models, and the highest balanced accuracy (0.77) was obtained with ROCKET combined with RidgeClassifierCV.

Aug 4, 2026

Upload by Noah