Identification of foot typology from gait analysis using machine learning approaches

Abstract

Human foot has unique anatomical and functional structure to facilitate various movements. It is difficult to classify the foot typology by visual assessment and by using static analysing tools. The aim of the study is to predict foot typology and classify the foot type using instrumental gait analysis and machine learning techniques, focusing on women aged 20-29 and 40-49 years to identify key features contributing to conditions like flat feet and high arch feet, and to facilitate early detection and correction of abnormal foot typologies. Data collection has involved 25 participants in Group 1 (aged 20-29 years) and 15 participants in Group 2 (aged 40-49 years), each undergoing 3 trials in instrumented treadmill gait analysis (ITGA), alongside questionnaires, consent forms, and body composition analysis. Participants have been selected based on specific inclusion and exclusion criteria to ensure valid results. The data has been used machine learning models to classify foot typology into Normal Arch, High Arch, and Flatfoot. The dataset has been separated as 70% training and 30% testing for multi-class classification. The foot typology has been classified using machine learning models based on 49 features of gait analysis. The features identify the patterns and differences among the different foot typologies for accurate classification. The application of synthetic minority over-sampling technique (SMOTE) to balance the dataset also improved the accuracy across all models. Among different machine learning models employed, bagging algorithm has achieved the highest accuracy of 95.6% and 94.3% for Group 1 and Group 2 respectively. The study has indicated effectiveness of machine learning techniques in classification of foot typology using gait analysis proving valuable insights into foot biomechanics and improved precision of diagnosis for personalized intervention strategies.