Text Mining Using Radial Basis Function Neural Networks and Optimization Algorithms رسالة ماجستير

Abstract

Text mining is a process for extracting information from an unstructured text. Text mining can work with unstructured or semi-structured data sets such as E-mails. Spam emails are unsolicited emails. It consumes storage of mail servers, waste of time and consumes network bandwidth. One of the most powerful tools that are used for email classification is Artificial Neural Networks (ANNs); it has the capability of dealing with huge amount of data with high dimensionality in better accuracy. In this thesis, we proposed a hybrid approach which combines Radial basis function Neural Network (RBFNN) and Particle Swam Optimization (PSO) algorithm. This approach is applied on two applications; the first is HRBFN-PSO that is applied on function approximation and time series prediction. The other is HC-RBFN-PSO which is used for classification in order to classify the emails spams. In the both proposed hybrid applications (HRBFN-PSO and HC-RBFN-PSO), the parameters of RBFNN are optimize as follow: Center is optimized using Particle Swam Optimization algorithm (PSO), Radii is optimized using K-Nearest Neighbors algorithm (KNN), and weights is optimized using Singular Value Decomposition algorithm (SVD). These two traditional algorithms (KNN and SVD) are integrated within each iterative process of Particle Swam Optimization, the weights and Radii are updated depending on the fitness (error) function. In the first application of the function approximation and time series prediction, the method HRBFN-PSO conducts experiments on nonlinear benchmark mathematical functions and Mackey Glass time series. The results obtained on the training data clarify that the proposed approach improved the approximation accuracy more than other traditional approaches. Also, this result shows that, HRBFN-PSO reduces the Root V Mean Square Error (RMSE) and Sum Square Error (SSE) dramatically compared with other approaches. Through our experiments for function approximation, we got the best RMSE value which is 0.000034. However, in the other application, the proposed method HC-RBFN-PSO conducted experiments on benchmark spam dataset namely SPAMBASE downloaded from UCI Machine Learning Repository, to classify emails into two classes namely spam and non-spam. The experimental results of this application show that our approach is more accurate compared with other approaches that use the same dataset. Through our experiments for email classification, we got the best classification accuracy value is 93.7% that is best compared with other approaches, they had obtained the best accuracy value is 93.28.