Classification of diabetic walking through machine learning: Survey targeting senior citizens









Abstract

We have an interest in diabetes, a metabolic disorder in which the levels of glucose in the blood are very high. Recently, the number of senior citizens who are detected with this disease is rapidly increasing. Moreover, diabetes does not end by lowering the levels of glucose concentration in the blood since it also causes different health complications while the disease is active, reducing the lifespan of patients. Thus, this study proposed a method to predict the possibility to find diabetes at its early stages through machine learning. The dataset for training consisted of nine features of senior citizens' walking data measured with the shoe-type IMU sensor at three different speeds (fast, slow and preferred) of 200 human subjects in their 60s-80s. With this, we created a program which is able to predict whether a patient has diabetes or not by using Machine Learning algorithms such as Logistic Regression, Support Vector Machine and Random Forest. We also compared the accuracies obtained for each algorithm and found that both Support Vector Machine and Logistic Regression models reached an 84% of accuracy. Through the analysis results, we determined the feature importance for learning, which showed high importance for fast walking features. It was discussed that this could be related to problems with diabetic plantar ulcers when patients suffered from diabetes.


Modules


Algorithms


Software And Hardware