SEDLF-LDD: A Stacking Ensemble-Based Deep Learning Framework for Lung Disease Diagnosis

Abstract

There is a growing need for accurate and swift diagnostic tools for lung disease diagnosis in healthcare. This work presents a Stacking Ensemble-based Deep Learning Framework for Enhanced Lung Disease Diagnosis (SEDLF-LDD). The stacking is a widely used ensemble learning technique that enhances the model's performance by combining the predictions from multiple base-learners using a meta-learner. The proposed framework selects the five best-performing pre-trained models, namely, ResNet50, MobileNetV2, VGG16, VGG19, and DenseNet201, as the base-learners and Multilayer Perceptron (MLP) as a meta-learner. To ensure broader applicability, we curated a dataset of chest X-ray images of Lung Disease. Initially, we choose the ten transfer learning models, fine-tune them to extract features relevant to respiratory diseases on the dataset, and select Top-5 best-performing models as base-learners. The effectiveness of the framework is determined by analysis of precision, recall, F1-score, or the area under the receiver operator characteristic (AUC-ROC) curve. The experimental results show an effective result with 97.65% accuracy.