FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W4386700634> ?p ?o ?g. }

• W4386700634 endingPage "S180" @default.
• W4386700634 startingPage "S179" @default.
• W4386700634 abstract "Machine learning (ML) methods, used in gait analysis, allow for accurate and quick analysis for applications like determination of ground contact timings (GCT) in healthy adults [1]. However, a known caveat of these methods is their inability to make accurate predictions on unseen datasets from different biomechanical activities due to different data characteristics [2]. In this study, we tested four supervised ML approaches trained on treadmill walking to identify GCTs for overground walking and jogging, to determine the best architecture for detecting GCTs across different gait conditions. How do ML architectures, trained on treadmill walking, perform on unseen gait conditions? 3-D kinematic and kinetic data were collected from 27 healthy adults (25±3.4 years). 3-D IMU signals (input) from sensors placed at the participant's waist and feet were sampled at 120 Hz. GCT was determined using in-shoe pressure insoles sampling at 100 Hz (ground truth). Each participant performed overground walking and jogging (self-selected speeds; 400 steps each), and treadmill walking (3.5-6 km/h; increments of 0.5 km/h; 600 steps). Four supervised ML models were implemented using treadmill walking data from 21 participants (Fig. 1). Algorithm performance was tested on treadmill walking, overground walking and overground jogging from the remaining 6 participants. A 5-fold cross validation was performed to test for overfitting. A kinematic method for detecting GCT was also tested to compare efficacy [3]. RMSE was calculated on the difference between the true GCTs and those predicted by each algorithm in milliseconds. Fig. 1: Schematic diagram of the methodology. 4 ML approaches were used to approximate target function f that mapped IMU signals from the waist and feet (y) to GCTs obtained from vGRF data (x). The implemented ML approaches were tested on overground walking and jogging. KNN- K-nearest neighbour; RF- Random Forest; SVM- Support Vector Machine; NN- Neural Network.Download : Download high-res image (72KB)Download : Download full-size image On treadmill data, RMSE±SD (ms) values of 8.1±3.3 (KNN), 9.2±3.8 (RF), 8.4±4.3 (SVM) and 6.9±3.8 (NN) were observed. For overground data, values of 14.3±5.7 (KNN), 18.2±7.2(RF), 15.3±4.6 (SVM) and 16.7±6.3 (NN) for walking, and 15.3±3.2 (KNN), 19.4±4.3 (RF), 16.1±3.6 (SVM) and 18.7±6.8 (NN) for jogging were observed. The kinematic algorithm reported values of 23.3±10.8 (overground walking), 21.8±7.5 (overground jogging) and 28.4±12.6 (treadmill walking). Outcomes suggest that while there isn’t a “one-model-fits-all” approach due to the inability of ML approaches to generalize to unseen gait conditions, all ML algorithms outperformed the kinematic algorithm, since kinematic algorithms rely on heuristics which may not hold true for biomechanically different activities like walking and jogging in different conditions. Overall, the KNN approach demonstrated the best results when tested on unseen data. Hence, our results suggest that simpler ML algorithms may have an advantage over NN approaches in this regard. However, a more comprehensive comparison of ML architectures is required before drawing a definitive conclusion." @default.
• W4386700634 created "2023-09-14" @default.
• W4386700634 creator A5006613317 @default.
• W4386700634 creator A5026498515 @default.
• W4386700634 creator A5064688639 @default.
• W4386700634 creator A5077861705 @default.
• W4386700634 creator A5092860695 @default.
• W4386700634 date "2023-09-01" @default.
• W4386700634 modified "2023-10-18" @default.
• W4386700634 title "A comparison of machine learning architectures for determining ground contact timings in overground and treadmill gait" @default.
• W4386700634 doi "https://doi.org/10.1016/j.gaitpost.2023.07.217" @default.
• W4386700634 hasPublicationYear "2023" @default.
• W4386700634 type Work @default.
• W4386700634 citedByCount "0" @default.
• W4386700634 crossrefType "journal-article" @default.
• W4386700634 hasAuthorship W4386700634A5006613317 @default.
• W4386700634 hasAuthorship W4386700634A5026498515 @default.
• W4386700634 hasAuthorship W4386700634A5064688639 @default.
• W4386700634 hasAuthorship W4386700634A5077861705 @default.
• W4386700634 hasAuthorship W4386700634A5092860695 @default.
• W4386700634 hasConcept C121332964 @default.
• W4386700634 hasConcept C12847509 @default.
• W4386700634 hasConcept C146849305 @default.
• W4386700634 hasConcept C151800584 @default.
• W4386700634 hasConcept C154945302 @default.
• W4386700634 hasConcept C1862650 @default.
• W4386700634 hasConcept C2781464450 @default.
• W4386700634 hasConcept C39920418 @default.
• W4386700634 hasConcept C41008148 @default.
• W4386700634 hasConcept C44154836 @default.
• W4386700634 hasConcept C70770792 @default.
• W4386700634 hasConcept C71924100 @default.
• W4386700634 hasConcept C74650414 @default.
• W4386700634 hasConcept C96332660 @default.
• W4386700634 hasConcept C99508421 @default.
• W4386700634 hasConceptScore W4386700634C121332964 @default.
• W4386700634 hasConceptScore W4386700634C12847509 @default.
• W4386700634 hasConceptScore W4386700634C146849305 @default.
• W4386700634 hasConceptScore W4386700634C151800584 @default.
• W4386700634 hasConceptScore W4386700634C154945302 @default.
• W4386700634 hasConceptScore W4386700634C1862650 @default.
• W4386700634 hasConceptScore W4386700634C2781464450 @default.
• W4386700634 hasConceptScore W4386700634C39920418 @default.
• W4386700634 hasConceptScore W4386700634C41008148 @default.
• W4386700634 hasConceptScore W4386700634C44154836 @default.
• W4386700634 hasConceptScore W4386700634C70770792 @default.
• W4386700634 hasConceptScore W4386700634C71924100 @default.
• W4386700634 hasConceptScore W4386700634C74650414 @default.
• W4386700634 hasConceptScore W4386700634C96332660 @default.
• W4386700634 hasConceptScore W4386700634C99508421 @default.
• W4386700634 hasLocation W43867006341 @default.
• W4386700634 hasOpenAccess W4386700634 @default.
• W4386700634 hasPrimaryLocation W43867006341 @default.
• W4386700634 hasRelatedWork W1967416733 @default.
• W4386700634 hasRelatedWork W1980265911 @default.
• W4386700634 hasRelatedWork W1989142914 @default.
• W4386700634 hasRelatedWork W2025046796 @default.
• W4386700634 hasRelatedWork W2030550205 @default.
• W4386700634 hasRelatedWork W2103412211 @default.
• W4386700634 hasRelatedWork W2379635356 @default.
• W4386700634 hasRelatedWork W2521657364 @default.
• W4386700634 hasRelatedWork W3112016203 @default.
• W4386700634 hasRelatedWork W4379410641 @default.
• W4386700634 hasVolume "106" @default.
• W4386700634 isParatext "false" @default.
• W4386700634 isRetracted "false" @default.
• W4386700634 workType "article" @default.