الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 136 |  |  |
| | | from | 136 | | |
Abstract
Chronic kidney disease (CKD) is still a health concern despite advances in surgical care and treatment. chronic kidney Disease is a fatal condition that is expanding quickly and day by day throughout the entire world it is one of the top 20 killers worldwide. According to the World Health Organization (WHO) chronic kidney disease has attracted a lot of interest as it have an increased problem due to its high rate of mortality. About 10% of the adult population globally has been diagnosed with chronic kidney disease and millions of people die each year as a result of their country’s economic situation. chronic kidney disease is characterized by a steady decline in renal function that can last for months or, if untreated, even years. CKD is a condition that impairs healthy kidney function. Failure of the kidneys can be avoided if chronic renal disease is detected and treated early. Early detection and treatment of chronic renal disease are the most effective methods. CKD’s growth in recent years has gained much interest from researchers around the world in developing high-performance methods for diagnosis, treatment and preventive therapy. Improved performance can be accomplished by learning the features that are in the concern of the problem. In addition to the clinical examination, analysis of the medical data for the patients can help the health care partners to predict the disease in early stage. In this thesis, we suggested a Deep Learning model based on modified Deep Belief Network (DBN) that works as a significant enabler for CKD. Several hidden layers are incorporated into the proposed DBN-based model to improve feature abstractions and functionality. The three stages that form it are preprocessing, feature selection, DBN training and detection phases. The first stage is preprocessing phase that handling the dataset which has more than half of the variables missing, making it necessary to handle those missing values in order to increase accuracy. When a value is missing, the attribute’s maximum frequency is utilized to fill in the blank using the Mode imputation approach. The normalizing of data scales to place all attributes into a certain range is another step in the preprocessing process. The second stage is the feature selection, in which reducing the elements that restrict effective processing by selecting the best features from all of the available features using the Density-Based Feature selection (DFS) algorithm in a wrapper approach that repeatedly applies the DFS technique in order to simplify and expedite the classification. Estimates of the classifier parameters are made by using the Grasshopper’s Optimization Algorithm (GOA) that estimate and fine-tune the classifier parameters. The application of optimization algorithms aids in parameter adaption to improve classifier performance. The third stage is classification that enables transferring the prepared data to an easy-to-use Deep Learning classifier we use the optimized deep belief network ODBN classifier that based on stacking Restricted Boltzmann Machine (RBM) to construct the deep network with the Softmax as activation function and the Categorical Cross-Entropy acting as a loss function. The proposed framework is evaluated using number of assessment indicators through extensive experiments using UCI benchmark dataset and compared with the most advanced methods. The trials’ outcomes demonstrate that they all outperformed their deep learning and machine learning equivalents, yielding notable performance benefits. The experimental study evaluated the suggested framework using a variety of accuracy measures and found that by achieving a high detection rate and they achieve the best trade-off. The results: with using the DBN, the dataset is randomly divided in two parts; the first part contains 70% of overall collection of data to train the model. The second part is used for testing and it contains the reminder of the dataset (30%).Six performance measures are used to evaluate and validate the proposed model. These measures are Accuracy, Precision, Recall, F-Measure, Root Mean Square Error (RMSE) and Mean Absolute Error (MAE). The DBN model demonstrated that 0.370000 and 0.482874 for Mean Absolute Error and RMSE respectively, also with using the second model DFS-ODBN, the results was 0.2100 and 0.1230 for Mean Absolute Error and RMSE respectively.”