Wrapper-Based Feature Selection to Classify Flatfoot Disease

Musculoskeletal disorders of the foot are a common complaint in the population. It has been found a flatfoot prevalence of 13.6% in young adults and a prevalence of 26.62% in adults between 42 and 91 years. Different non-invasive techniques can identify the type of foot by analyzing the soles of the feet, such as the Chippaux-Smirak index (CSI). Although CSI is a non-invasive technique, it is performed manually, and the intervention of an expert is necessary to give a clinical opinion. The use of automatic systems is an alternative. This article introduces a machine learning-based tool that permits the identification of foot types. The proposal employs a wrapper feature selection mechanism to select the subset of features that improves the classification. This task is considered from an optimization perspective, and the optimal subset is chosen using metaheuristic algorithms. Eight algorithms used in the optimization are compared, and an increase in the Accuracy of the K-nearest neighbors (KNN) classifier is observed from 73.5% to 94.7%. Of the 39 total features proposed in the dataset, only 10 features are considered significant. The significance of the characteristics implies that they have an effect on the morphology of the foot. If they are considered in treatments to minimize this disease, it can reduce their development costs.