XceptRf-Net: A Novel Deep Learning and Machine Learning Approach for Pneumonia Diagnosis

The objective of this study is to develop an advanced and interpretable diagnostic framework combining deep learning and machine learning for the initial diagnosis of pneumonia with high accuracy in pediatric patients to overcome the critical limitations of existing diagnostic procedures.

We introduce a hybrid model, XceptRF-Net, that integrates deep feature learning of the Xception convolutional neural network and the probabilistic modelling power of the Random Forest for the classification of Fisher’s iris dataset. The first stage of the model considers the highlevel spatial features from the chest X-rays, which are extracted using Xception. Followed by that, they are subsequently mapped to a probabilistic feature space using Random Forest, contributing to the feature representation and the classification robustness. The discriminative capability of the engineered features was tested by different machine learning classifiers such as Logistic Regression (LR), K-Nearest Neighbours (KNN), and Multi-Layer Perceptron (MLP). Fine-tuning and k-fold cross-validation were also performed for generalization purposes and to speed up computation.

The proposed XceptRF-Net framework, established from an experimental study on one dataset with 5,863 pediatric CXRs, has been objectively shown to benefit over conventional methods. Logistic Regression achieved the highest diagnostic accuracy of 98%, which is a validation of the spatial and probabilistic feature learning integration.

The effectiveness of the XceptRF-Net model highlights the value of combining deep feature extraction with probabilistic modeling to enhance clinical decision-making.

The findings highlight the theoretical superiority of the integration of convolutional deep features with ensemble learning and the generation of probabilistic features for medical image analysis. The proposed method provides a stable and explainable framework for clinical decision support and has high potential for practical use in real-world systems for pediatric pneumonia screening and diagnosis.