clustering of pelvic-supporting muscles activity in individuals with lumbar hyperlordosis in stance phase of walking: a machine learning approach

Introduction: lumbar hyperlordosis (lh) is associated with lumbar muscle defects or altered muscle engagement patterns, leading to low back pain. however, the specific muscle most influential in causing this condition remains unclear. objectives: this study aims to determine effective prescient alternative to lumbar hyperlordosis (lh) from muscular activity variables in stance phase of walking, in addition to discovering homogenous clusters of individuals based on the primary predictive alternative. material and methods: the activity of rectus femoris (rf), gluteus medius (gm), and lumbar erector spinae (les) was recorded in 40 females suffering from lh while walking. maximum activity and muscular involvement for each muscle were extracted. a multilayer perceptron artificial neural network was used to detect notable projected variables of lh. k-means clustering was then employed to identify homogeneous clusters of individuals based on the most significant predictive variable. the one-way anova test used to identify homogenous clusters. results: the results demonstrated that rf maximum activity with an accuracy of 90.9%, was detected as the most prominent predictive variable. the one-way anova test demonstrated significant differences among the three homogeneous clusters of individuals based on rectus femoris maximum activity (p≤0.05). conclusion: the classification scheme presented in this paper can describe muscle activity patterns while walking and may be useful for screening individuals suffering from lumbar hyperlordosis and for clinical decision-making based on clusters. the maximum activity of the rectus femoris is the most important factor affecting lumbar hyperlordosis, which is relevant in rehabilitation and health fields.