Cancer Classification using improved Extreme Learning Machine
Ankita Shreya, Swati Vipsita, Santos Kumar Baliarsingh · 2019
In recent years there has been explosion in the amount and complexity of micro-array datasets obtained from various biological experiments and many community research. These datasets when utilized efficiently can help in building of complex disease prognosis systems. Efficient and accurate classification model is needed for analysis and interpretation of these datasets. Extreme Learning Machine (ELM) has been catering the attention for training single-layer feed forward network (SLFN) by providing higher performance in terms of fast learning speed and better accuracy. The efficiency of these models are heavily dependent on the selection of network parameters which involves hidden weights and biases. Therefore, many evolutionary algorithms have been used to enhance the performance of ELM. However, these evolutionary approaches are limited to explore the global search space, i.e. diversification. In order to obtain high quality solution, along with diversification searching the local neighbourhood i.e, intensification is required. In the proposed work, a novel method is developed where Jaya optimization algorithm and Simulated Annealing (SA) is together implemented to produce a profitable synergy. Further, adaptive technique is introduced to enhance the search process. Comparison and experiments on five micro-array datasets reveals the robustness of the proposed ELM models. The experimental results demonstrate that the proposed hybridized ELM are better than the original ELM algorithm and other existing state-of-art in terms of accuracy and efficiency.