الفهرس 
Chapter1: INTRODUCTIONOnly 14 pages are availabe for public view
 |  | from | 333 |  |  |
| | | from | 333 | | |
Abstract
Medical data analysis plays a crucial role in the bioinformatics field, enabling predictions of diseases and suitable drugs. Biomedical informatics, a branch of health informatics, leverages data to improve human health and healthcare. While many successful medical applications aid in disease diagnosis, current research mostly relies on lab experiments for data collection. In this thesis, we utilize real-world medical data from multiple resources and apply soft computing techniques to enhance disease and drug prediction.
The main idea focuses on using several medical data sets to diagnose diseases or propose the best drug. Thus, we deal with the data to produce accurate models. Moreover, to solve high-dimensional data problems, we remove unnecessary data from a data set to determine the most important features and decrease the dimensionality of the data set. FS also improves classification accuracy and decreases model complexity. The FS problem can be considered an optimization problem, so using optimization algorithms to perform the critical task accurately is recommended. Optimization algorithms may integrate with SVM classifier to propose the optimal diagnose or drug. Also, optimization algorithms are used to optimize the most important variables in training the classifier, which are the connection weights and biases, so optimization algorithms may be used to enhance the FNNs classification efficiency. All the above points are critical situations in several applications. Moreover, propose a new model to optimize FS or the FNNs that classify different medical data.
Thesis contributions
The main contributions of this thesis are listed in brief as follows:
1- Presents a comprehensive review to study state-of-the-art challenges and recommended techniques for medical data classification. Provides ML techniques that are currently applied to classifying medical data using various data sets. Illustrates several bioinformatics computational methods used for medical data classification.
2- Proposes an enhanced version of a meta-heuristics algorithm called the modified Hunger Games Search Algorithm (mHGS), for solving optimization and FS problems. The traditional HGS is enhanced by adding the following mechanisms: fuzzy logic-based mutation for control parameters, balancing the strategy of exploration and exploitation, and a population reduction strategy. mHGS is proposed to improve its local search capability and solve the problem of premature convergence. mHGS is proposed as an alternate feature selection approach. The SVM method was used for classifying the data, and it had a high average accuracy rate of 98.060\%. Furthermore, using the mHGS for FS significantly improved the SVM classification performance. The experimental results demonstrated that the proposed method outperformed others in classification.
3- Proposes a wrapper FS approach that hybridizes the rat swarm optimization (RSO) algorithm with genetic operators with time-varying transfer functions (TFs) to control local and global search and avoid local solutions. First, RSO was boosted using genetic operators to avoid local optimality. Then, by converting continuous search spaces to discrete search spaces, transfer functions are used to balance local and global search spaces. Second, the proposed model was applied to classification using SVM to increase accuracy and decrease the number of features. Furthermore, eight chemical data sets and four toxicity effect data sets were used to test the proposed algorithm for FS over real data sets. Then the bmRSO is hybridized with the SVM as a classifier process (called bmRSO-SVM) for classification purposes. bmRSO1-SVM achieves optimal results of accuracy 98.201% compared with other versions and state-of-the-art algorithms.
4- Moreover, it suggests a modified weighted mean of vectors algorithm called (mINFO) that combines INFO and the enhanced solution quality (ESQ) operator. By avoiding ideal local values throughout each iteration, confirming that each solution advances towards a better location, and speeding convergence, the ESQ is used to enhance the quality of solutions. Moreover, the proposed mINFO method was applied for training FNNs and multi-layer perceptrons to increase their accuracy. mINFO proposes the best results for ten chemical data sets compared with other algorithms.