الفهرس | Only 14 pages are availabe for public view |
Abstract Cancer causes abnormal growth in cells and sometimes it leads to death. Breast cancer is one of the most common types of cancer. It can occur in both men and women, but it’s far more common in women. Detection of cancer in its early stages and treatment can greatly improve the survival rate of patients. To know if the breast cells have cancer or not, seven contributions are presented in this thesis based on Wisconsin Diagnostic Breast Cancer (WDBC) dataset. The first contribution is improving the performance of the classification algorithm Support Vector Machine (SVM) using a recent Harris Hawks Optimization (HHO) to diagnose breast cancer (HHO-SVM). HHO-SVM achieves 98.24% against 96.66% accuracy rates for the SVM classifier with Grid Search Algorithm (Grid-SVM). The second contribution is improving the performance of SVM using a recent Grey Wolf Optimization (GWO) to diagnose breast cancer (GWO-SVM). GWO-SVM achieves 98.60% against 96.66% accuracy rates for Grid-SVM. Ten scaling techniques were efficient for linear programming. These scaling techniques are applied with SVM on the WDBC dataset. They are arithmetic mean, de Buchet for three cases (𝒑=𝟏,𝟐,∞), equilibration, geometric mean, IBM MPSX, Lp-norm for three cases (𝒑=𝟏,𝟐,∞). By applying these scaling techniques, we have three other contributions, and they are as follows: The third contribution is improving the performance of Abstract iv Grid-SVM from 96.66% to 98.95% accuracy rate. The fourth contribution is improving the performance of GWO-SVM from 98.60% to 99.30% accuracy rate. The Fifth contribution is improving the performance of HHO-SVM from 98.24% to 99.47% accuracy rate. CPU time of HHO-SVM and GWO-SVM models are large so, the sixth contribution is proposing the parallel version of the GWO-SVM achieves a speedup by a factor of 3.91 on four cores. Finally, the seventh contribution is the parallel version of the HHO-SVM achieves a speedup by a factor of 3.97 on four cores. |