Turkish Journal of Physiotherapy and Rehabilitation ; 32(2) ISSN 2651-4451 | e-ISSN 2651-446X www.turkjphysiotherrehabil.org 2993 ADAPTIVE SPARSE K-MEANS AND OPTIMIZATION ENABLED NEURAL NETWORK FROM GENE-EXPRESSION DATA FOR CANCER CLASSIFICATION LAKSHMI RAMANI BURRA 1 , BHRAMARAMBA RAVI 2 , BANOTHU RAMJI 3 , APPARNA ALLADA 4 , PRAVEEN TUMULURU 5 1 Assistant Professor, Dept. of CSE, PVP Siddhartha Institute of Technology, Vijayawada, ramanimythili@gmail.com 2 Professor, Dept. of CSE, GIT, GITAM Deemed to be University, Visakhapatnam, bhramarambaravi@gmail.com 3 Assistant Professor, Dept. of CSE, CMR Technical campus, Hyderabad, Telangana, vasramji@gmail.com 4 Assistant Professor, Dept. of CSE, Gudlavalleru Engineering College, Gudlavalleru, alladaaparna9@gmail.com 5 Assistant Professor, Dept. of CSE, Koneru Lakshmaiah Education Foundation, Vaddeswaram, praveenluru@gmail.com ABSTRACT Cancer is one of the malignant diseases existing globally and the people affected with cancer are rescued only when the disease is recognized at the earliest possible stage. Identify in advance of disease is essential as in the final stage; since the chance of living/existence is partial. The indications of cancers are difficult and thus, all the indications should be considered accurately earlier to the diagnosis. Thus, an automatic prediction system is essential for classifying the tumor to malignant or benign. This work introduces a cancer classification approach using Chronological Grasshopper Optimization Algorithm (Chronological-GOA) for classification of cancer. For reducing the dimension of gene-expression data, log transformation is applied to the database. Then, the adaptive sparse K-means clustering selects the necessary data, which is provided to the Deep Belief Network (DBN). Here, the DBN is trained using Chronological-GOA. At last, the DBN classifies the selected gene sequences as normal and abnormal gene, and thereby identify the cancer. The performance of the cancer classification based on MSparse Kmeans + Chronological GOA-DBN is computed based on accuracy, detection rate, and False Alarm Rate (FAR). The developed method attains the accuracy of 0.9876, maximal detection rate of 0.9893, and the minimal FAR of 0.0596. Keywords: Gene-expression data, Adaptive sparse K-means clustering, Chronological GOA, Deep belief network, log transformation. I. INTRODUCTION One of the dangerous diseases caused by most of the living organisms is cancer. Research on Cancers has been paid more attention in the field of medicine. With the rapid growth of microarray approach, the tumor classification research has been paid novel breakthrough in the recent years [4]. In the last few years, the mortality rate of cancer has been grown rapidly, causing a serious problem to human health. Uncontrolled proliferation and metastasis of cancer cells is challenging due to the identification of other types of cancer. Most of the cancers are diagnosed at the very initial stage [1], but becomes deadly at the final stage. Cancer classification is utilized for enhancing the health-care of patients, and life quality of the individuals, and, also it is suitable for drug discovery, and the diagnosis of cancer [2] [8] [29]. Gene is nothing, but the physical and functional unit of heredity and genes are made by Deoxyribonucleic acid (DNA). Few genes are act as instructions for making molecules termed proteins [3]. The gene expression level is utilized to solve the fundamental issues related to biological evolution mechanisms, drug discovery, prevention