الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 130 |  |  |
| | | from | 130 | | |
Abstract
Nowadays, Bioinformatics is a very critical and important field as it is related to human health like diagnosing diseases and molecular biology. Therefore, many governments have become interested in developing bioinformatics as an important disciplinary in its universities.
Cardiotocography is one of the most common medical devices. It is utilized in monitoring fetal heart rate and uterine contraction during the period of pregnancy. Doctors use cardiotocography to diagnose and classify a fetus state. Unfortunately, they have challenges in uncertainty of data. The Cardiotocography dataset is downloaded from University of California, Irvine (UCI) machine learning repository contains 2126 medical case. It is composed of 21 input attributes used to determine the state of fetal heart rate and uterine contraction. Depending on the input attributes values gynecologists could classify the state of fetal as normal, pathologic or suspicious state (NSP) class.
Many machine-learning scientists and researchers presented studies for dealing with the uncertainty and ambiguity of medical data generally and in cardiotocography data especially. They aimed to analyze and classify data in efficiently and to achieve good performance measurements like accuracy rate, precision, and recall. The automatic classification helps doctors prioritize critical cases that need to be quickly intervened to save them, which through self-diagnosis would put these cases at the back of doctor concerns compared with other more stable cases.
This thesis presents a proposed model called the Rough Neural Network model to classify the cardiotocography data. It utilized the Rough Neural Network technique, which is one of the most common machine learning techniques used to classify medical data. Moreover, it is a good solution for the uncertainty challenges. The proposed model measured each of the accuracy rates and consumed time during the classification process. During Experiments, WEKA mining tool is utilized to analyze cardiotocography data with different algorithms (neural network, decision table, bagging, the nearest neighbor, decision stump, and least square support vector machine algorithm). This comparison showed that the accuracy rate and time consumption of the proposed model was more feasible and efficient than other techniques.
The rough neural network model is very efficient in handling uncertainty, but not indeterminacy. This thesis also enhanced the proposed RNN model by applying neutrosophic set theory to manage the degree of indeterminacy in the classification decision. The second proposed neutrosophic diagnostic model is an Interval Neutrosophic Rough Neural Network that is based on back-propagation algorithm. It benefits from the advantages of neutrosophic set theory to improve the performance of rough neural networks and satisfy a better performance than other literature algorithms. The experimental results visualized the data using the boxplot for a better understanding of attributes distribution. The performance measurements of the confusion matrix for the proposed model are 95.1, 94.95, 95.2, and 95.1 for accuracy rate, precision, recall, and F1-score respectively. In addition, the Receiver Operating characteristic Curve (ROC) indicated that the proposed model classified normal, suspicious, and pathologic states by 0.93, 0.90, and 0.85 areas under the curve respectively that are considered high and acceptable.