Comparative Analysis of Hybrid and Ensemble Learning in Lung Cancer Diagnosis

Abstract

Lung cancer remains one of the leading causes of cancer-related deaths globally, primarily due to delayed diagnosis. Early and accurate detection is critical for improving patient survival rates. However, microarray data used in cancer diagnosis pose a significant challenge due to the curse of dimensionality, where a small number of samples are associated with a large number of features, potentially reducing the accuracy (ACC) of prediction models. To address this, we propose a hybrid machine learning (ML) approach that integrates correlation-based feature selection (CFS) and the elephant search algorithm (ESA) for effective feature selection and optimization. Additionally, we introduce an ensemble deep learning (EDL) strategy, combining a deep neural network (DNN) with a bagging classifier and ensemble techniques such as weighted averaging (WA), soft voting (SV), and hard voting. Experiments conducted on a microarray dataset from the national center for biotechnology information (NCBI) evaluated performance using ten metrics. The hybrid approach (CFS+ESA+MLP) achieved an ACC of 97.18%, while the ensemble DL model with hard voting (HV) attained a superior ACC of 98.87%. These results demonstrate the effectiveness of the proposed methodologies in enhancing early lung cancer diagnosis.