Automated Pneumonia Detection Using a Custom Convolutional Neural Network: A Balanced Approach

Pneumonia is a severe respiratory infection that presents considerable health hazards globally, particularly in developing nations where insufficient medical resources sometimes impede early detection. Recent advancements in deep learning, particularly convolutional neural networks, have demonstrated significant potential in the automated diagnosis of pneumonia from chest X-rays. This could be a cost-effective and accurate solution for healthcare settings that do not have many resources. In order to accurately classify pneumonia cases, this paper proposes a custom CNN model that addresses common issues such as class imbalance and overfitting. The study fixed the problem of class imbalance by making the majority class (PNEUMONIA) the same size as the minority class (NORMAL). This ensured that the training dataset was balanced, which helped the model generalize. Our suggested model has a sequential CNN architecture with four Conv layers, including ReLU activation, batch normalization, and MaxPooling layers. These are followed by fully connected layers, making feature extraction and classification work well. Dropout layers are applied in the deeper Conv and fully connected layers to prevent overfitting, ensuring the model generalizes well to unseen data. The Adam optimizer enhances the model via a binary cross-entropy loss function, while EarlyStopping and ModelCheckpoint oversee training to retain the optimal model, reducing overfitting and improving performance. The model underwent training on a balanced dataset for 20 epochs, with a maximum training accuracy of 98.03% and a validation accuracy of 99.78%. This demonstrates a significant improvement over existing models in the literature, highlighting the robustness and reliability of our approach.