The Effectiveness of Distinctive Information for Cancer Cell Analysis Through Big Data

Abstract

In the healthcare system, a huge volume of multi-structured patient data is generated from in-hospital clinical examinations, wearable body sensors, and doctor memos. These data play a deterministic role in finding the patient’s cause of disease and corresponding cure. However, there are many challenges underlying healthcare system such as integration of different data format, appropriate selection of healthcare parameters and disease prediction. Besides these, concern whether this data is equally significant for decision making or not exists. To address these problems, we introduced the infinite latent feature selection (ILFS) to find highly informative gene from a gene pool and feed it to the classifier. We adopted a hybrid approach for gene classification. The hybrid scheme integrated two (sophisticated) algorithms: speed advantage of extreme learning machine (ELM) and accuracy advantage of sparse representation classifier (SRC). We validated our proposed model with 198 tumor samples of global cancer map (GCM) dataset and divided them into 14 common tumor categories having 11,370 expression level of genes. The performance of this system was measured using five different matrices (i.e., accuracy, sensitivity, specificity, precision, and F-score). This achieved a notable improvement of accuracies in both scenarios in selective (considering only highly expressive genes) and original genes (74.75% and 81.82% respectively).