A hybrid feature selection and generative adversarial network for lung and uterus cancer prediction with big data

Abstract

Among all diseases affecting humanity, lung cancer has consistently stood out as one of the deadliest. It ranks among the most prevalent cancers and is a significant contributor to cancer-related deaths. The disease is often asymptomatic in its early stages, making early detection extremely challenging. To enhance the accuracy of cancer detection with minimal time, an effective hybrid feature selection and classification model is developed in this research for the efficient detection of detect lung and uterus cancers while leveraging big data. The Piecewise Adaptive Weighted Smoothing-based Multivariate Rosenthal Correlative Target Projection (PAWS-MRCTP) comprises three main processes namely data acquisition, preprocessing, and feature extraction.  In the data acquisition phase, a large number of cancer patient data are collected from lung cancer and uterus cancer detection datasets. Subsequently, the collected patient data undergo preprocessing. The preprocessing stage comprises three key processes namely handling missing data, noisy data, and outlier data. Firstly, the proposed PAWS-MRCTP is employed to address missing values, utilizing the Piecewise Adaptive Constant Interpolation method based on multiple available data points. Noisy data are identified using Gower's weighted smoothing technique, which detects data containing random variations or errors. Then the Improved Particle Swarm Optimization (IPSO) with fuzzy possibility C-Means clustering (FPCM) is introduced for the data clustering. And then the hybrid feature selection is performed using the ANFIS and Modified Chicken Swarm Optimization (MCSO). Finally, the classification of uterine and lung tumors is done using the Generative Adversarial Network (GAN). Consequently, in the experiments, the proposed model beats existing classifiers in detection accuracy while consuming the least time.