الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 109 |  |  |
| | | from | 109 | | |
Abstract
Recommender systems are software applications that aim to support users in their decision-making while interacting with large information spaces. Advanced recommendation models usually deal with the common challenges of enormous information (scalability) and lack of information (sparsity). The proposed recommender system uses rating prediction based on machine learning (ML) along with optimization algorithms in order to provide a solution for sparsity and scalability problems.
On other hand, disabilities, especially the ones caused by spinal cord injuries (SCI), affect both people’s behavior and participation in daily activities. So, people with SCI need long care, cost, and time to improve their health status. The proposed recommender system has been tested and validated via providing recommendations of the rehabilitation methods for patients with SCI. The predicted rehabilitation methods are provided via monitoring and recording the progress in patient’s health status over different periods of time. Accordingly, a set of rehabilitation methods has resulted based on prediction of user ratings.
The proposed recommender system is divided into four phases: preprocessing, clustering, recommendations, and prediction phases; during the preprocessing phase a SCI automated tool has been built to collect data of patients with SCI. Experimental results shows that the proposed SCI automated tool has an efficiency of 100%. Also, the rehabilitation length of stay (LoS) for patient with SCI is predicted using support vector machine; the accuracy is measured for linear and radial basis function (RBF) kernel functions, and the accuracy of linear function is totally match for training and test data, and 93.3% match for RBF kernel function. The proposed recommender system is conducting a comparative analysis between three proposed scenarios for rating prediction and the experimental results shows that the rating prediction using grey wolf optimization with Neural Network (GWO-NN) gives a better performance with 95% of accuracy and 92% of precision. Also, the accuracy and precision measures have been calculated to evaluate the overall system performance. Experimental results showed that the proposed recommender system has out performed other tested traditional recommender systems.
Some of machine learning such as clustering, classification, and optimization algorithms is combined with recommender system to get a better accuracy and performance. The proposed recommender system is based on rating prediction using three proposed scenarios: rating prediction using k-Nearest Neighbor (k-NN), rating prediction using GWO-NN, and rating prediction using sparse singular value decomposition (Sparse SVD).