الفهرس | Only 14 pages are availabe for public view |
Abstract his thesis presents intelligent computer models for analyzing COVID19 disease. These models help to classify cases of COVID-19 into COVID-19 and Non-COVID with high performance. Data mining and Machine Learning (ML) techniques are strongly recommended in developing these expert COVID-19 models which assist physicians in diagnosing and predicting COVID-19 disease in early stages. In this thesis, performance of several COVID-19 imaging techniques are discussed such as: Reverse transcription polymerase chain reaction (RTPCR), X-ray and Computed Tomography (CT). CT technique is one of modern imaging techniques that takes into account physiological alterations specific to characteristic of COVID-19. For analysis of COVID-19 in CT images, this thesis presents an intelligence model which identify positive COVID-19 cases. It presented the pipeline of medicinal imaging and examination methods included COVID-19 image acquirement, segmentation and diagnosis, using CT images. This thesis presented two effective models for single machine learning (SML) and ensemble machine learning (EML) to detect cases of COVID-19;The first classification model (SML) was applied with several algorithms, such as Decision Tree (DT), Artificial Neural Networks (ANN) and Support Vector Machines (SVM). Results showed that performance of SVM surpassed other classifiers with a 98.58 % accuracy. The second classification model (EML) was applied with several algorithms, such as Random Forest (RF), Voting and Bagging, to increase its accuracy up to 99.60 %, especially using Bagging classifier. Finally, the results of two proposed models outperformed those of other previous studies. EML, on other hand, performed even better than SML and is suggested for real-time application. This thesis also used effective feature selection (FS) algorithm for COVID-19 detection model, called PSO-FS algorithm. This proposed algorithm used particle swarm optimization (PSO) technique as FS search method for classification COVID-19. The proposed PSO-FS algorithm employs PSO algorithm to find significant and effective features subset within overall features set. Support Vector Machine (SVM), K-nearest neighbor (KNN) classifiers were used as evaluators. The accuracy reached 99.67% for SVM and 94.27 % for KNN respectively. Experimental results show that the proposed PSO-FS algorithm outperforms the other two traditional FS search methods, Genetic Algorithm (GA) and Greedy Stepwise (GS). This model is an intelligent and comprehensible model for medical experts. |