Leveraging Enhanced PSO and Proving Random Forest's Dominance for Prediction of Lung Cancer Severity

Abstract

One of the most fatal malignancies in the world is lung cancer. Usually, it starts in the cells lining the airways of the lung tissues. Close to 85% of lung cancer cases are caused by smoking, and hence making it the primary reason of the disease. However, lung cancer can also strike non-smokers, or passive smokers, for a variety of reasons, including genetics, asbestos exposure, radon gas exposure, and second hand smoke. Frequent signs and symptoms include a chronic cough, chest discomfort, sudden weight loss, dyspnea, etc. The high death rate linked to delayed diagnosis of lung cancer makes the need for machine learning algorithms to anticipate the disease more and more obvious. The lung cancer usually shows no symptoms (asymptomatic) till it progresses to an advanced stage. Hence, early identification and diagnosis become critical to increase the survival rates. Huge datasets, such as genetic profiles of patients, their histories, and medical pictures may be processed and analysed using the machine learning to find the patterns and risk factors that primitive approaches might miss. Timely identification made possible by these models allows for prompt treatments and more individualized treatment regimens. In this work, Enhanced Particle Swarm Optimization (PSO) is used to Pre-Process the data. It draws inspiration from collective behavior of swarm and uses Interval-Newton method. We used machine learning algorithms such as Random Forest, K-Nearest Neighbors, Multinomial Logistic Regression, and Support Vector machines in this paper to analyse the available data. We found that Random Forest performed better than other methods with 97% for Low, 94% for Medium, and 98% for High severity.