Stacked Autoencoder-based ELM with Unsupervised Feature Extraction for Efficient Classification of Tumors

Abstract

Abstract: Tumors may exist in the brain, lungs, esophagus, leukemia, breast, ovary and bladder. Each tumor has its pathogenesis and categorizing them is an uphill task. Appropriate features are to be extracted, and the data of tumor should be classified either as benign or malignant. This is a critical issue as treatment of patients is solely based on this. In this paper, tumor data is classified using Stacked Extreme Learning Machine (S-ELM) and features are extracted using Statistically Controlled Activation Weight Initialization (SCAWI). It involves unsupervised clustering and is based on Neural Network (NN) framework. The proposed method offers a good decision support system for classifying tumor data. The proposed scheme is applied to data obtained from hospitals of repute and publicly available domains and tested for its outperformance. Results are compared with basic ELM and Multilayer-ELM (ML-ELM). It is seen that S-ELM with SCAWI offers better performance in contrast to diverse methods of classification.

Index Terms: Breast Tumor, ELM, Multilayer-ELM, Tumor Classification.