ايات ناجي حسين
In many domains, gait rehabilitation is an interdisciplinary subject, which can be utilized in sports and health diagnostics since it can measure, characterize and assess many time and space characteristics. Gait rehabilitation can be utilized in stroke rehabilitation therapy to measure the patient’s level of rehabilitation and offer a quantitative reference for the physician to build a tailored walking training program for the patient. This Thesis proposes an intelligent system based on machine learning techniques to assist clinicians by evaluating enormous volumes of data and ensuring a thorough understanding of patient health records. Furthermore, the use of machine learning classifiers greatly aids in the speed and accuracy of diagnosis because machine learning algorithms work on the principle of training. This means their ability to predict disease based on previous data for patients with the same disease, which aids in the speed and ease of diagnosis. The proposed system includes three main phases: preprocessing phase, feature reduction phase, and classification phase. In the first phase, the bandpass filter was used in order to filter input data. Then the resulting data from the previous phase will be reduced using two feature reduction methods. Finally, six Machine Learning (ML) techniques were used to classify patients from healthy people. The proposed system has been tested using two datasets, one of the two datasets is available on the internet and the other is collected from the hospital. The first data contain four classes: healthy, mild, moderate, and severe. This dataset contains information about 169220 patients. The other dataset was collected by using the sEMG device from Foot Drop Patients in Metro Rehabilitation Hospital in Sydney, Australia using Ethical Approval (UTS HREC NO. ETH15-0152). Six classifiers were used for the purpose of helping to diagnose foot diseases and speed I up leg rehabilitation. A set of operations were applied to the data, the most important and most influential was the use of Principal Component Analysis (PCA) and Singular Value Decomposition (SVD) techniques to extract and reduce features. The results obtained indicate that the classifiers K-Nearest Neighbors (KNN) and Random Forest (RF) were the best for the first dataset, where the classification accuracy rate in both of them was 99% while the Naïve Bayes (NB) returned the lowest accuracy rate equal 78%. As for the second dataset, the classifiers K-Nearest Neighbors (KNN), Random Forest (RF), Decision Tree (DR), and Naïve Bayes (NB) shared the highest classification accuracy, which is 100% while the Stochastic Gradient Descent (SGD) and Logistic Regression (LR) returned the lowest accuracy rate equal 98% for both of them.
