Strategies to optimise machine learning classification performance when using biomechanical features - Article

Aug 4, 2026

Upload by Aurora

Building prediction models using biomechanical features is challenging because such models often need large sample sizes, yet collecting biomechanical data at scale is difficult. This study tests whether modern machine learning can mitigate limited sample sizes using secondary analysis of two datasets: a walking dataset (2295 participants) and a countermovement jump dataset (31 participants). Using 3D ground reaction forces, multiple models (regression, XGBoost, deep time-series with augmentation, and transfer learning) were compared, with AUC gains under data augmentation and transfer learning.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages7

Reading Time18 min

AccessFree to download

Introduction

Predictive modeling and sample-size challenge

Machine learning and deep learning in small-data regimes

Transfer learning and data augmentation

Journal of Biomechanics 165 (2024) 111998 Contents lists available at ScienceDirect Journal of Biomechanics journal [homepage: www.elsevier.com/locate/jbiomech](homepage: www.elsevier.com/locate/jbiomech) | Strategies to optimise machine learning classification performance when using biomechanical features
Bernard X.W. Liewa, *, Florian Pfistererb, c, David Rügamerb, c, Xiaojun Zhaid
a School of Sport, Rehabilitation and Exercise Sciences, University of Essex, Colchester, Essex, United Kingdom b Department of Statistics, LMU Munich, Munich Germany
c Munich Center for Machine Learning, Munich, Germany
d School of Computer Science and Electrical Engineering, University of Essex, Colchester, Essex, United Kingdom | | | | --- | --- | --- | | A R T I C L E I N F O | | A B S T R A C T | | Keywords: Machine learning Deep learning Gait Biomechanics Orthopedic
Musculoskeletal pain | | Building prediction models using biomechanical features is challenging because such models may require large sample sizes. However, collecting biomechanical data on large sample sizes is logistically very challenging. This study aims to investigate if modern machine learning algorithms can help overcome the issue of limited sample sizes on developing prediction models. This was a secondary data analysis two biomechanical datasets – a walking dataset on 2295 participants, and a countermovement jump dataset on 31 participants. The input features were the three-dimensional ground reaction forces (GRFs) of the lower limbs. The outcome was the orthopaedic disease category (healthy, calcaneus, ankle, knee, hip) in the walking dataset, and healthy vs people with patellofemoral pain syndrome in the jump dataset. Different algorithms were compared: multinomial/ LASSO regression, XGBoost, various deep learning time-series algorithms with augmented data, and with transfer learning. For the outcome of weighted multiclass area under the receiver operating curve (AUC) in the walking dataset, the three models with the best performance were InceptionTime with x12 augmented data (0.810), XGBoost (0.804), and multinomial logistic regression (0.800). For the jump dataset, the top three models with the highest AUC were the LASSO (1.00), InceptionTime with x8 augmentation (0.750), and transfer learning (0.653). Machine-learning based strategies for managing the challenging issue of limited sample size for biomechanical ML-based problems, could benefit the development of alternative prediction models in healthcare, especially when time-series data are involved. | 1. Introduction Gait impairments are common in many orthopedic (Biggs et al., 2022), musculoskeletal (Diamond et al., 2017), neurological (de Freitas Guardini et al., 2021), and cardiovascular disorders (Green et al., 2016). The quantification of gait impairments for use in predictive models can serve in facilitating clinical decision-making (Chia et al., 2020), and stratify patients to homogenous functional severity levels (Tsitlakidiset al., 2019) for allocation resourcing, and prognostication (Capin et al., 2017; de Freitas Guardini et al., 2021). Predictive models are typically required to understand the relationship between a set ofrisk/prognostic factors and clinically relevant outcomes (Shibuya et al., 2020). A challenge in the development of predictive models is the issue of sample size. For example, using 10, 20, and 50 events per predictor parameter rule (Cruz et al., 2020; Riley et al., 2019), for just 20 included predictors, the number of required participants will exceed some of the largest prospective clinical cohort studies to date (n = 2758 participants (Traeger et al., 2016)). While new techniques are emerging quickly in machine learning (ML) and deep learning, many studies show that tree-based gradient boosting techniques such as XGBoost (Chen and Guestrin, 2016) still outperform most techniques, especially, when the sample size is small (Benkendorf and Hawkins, 2020

FAQs

Why are biomechanical-feature prediction models difficult to build?

They may require large sample sizes, while collecting biomechanical data at scale is logistically challenging.

What datasets and inputs were used in the study?

Two secondary datasets were analyzed: a walking dataset (2295 participants) and a countermovement jump dataset (31 participants), with 3D ground reaction forces as input features.

Which modeling strategies were compared to address limited sample size?

Multinomial/LASSO regression, XGBoost, deep learning time-series methods with augmented data, and transfer learning were compared using AUC-based outcomes.

Aug 4, 2026

Upload by Aurora