Optimizing Healthcare With Machine Learning: Protecting Finances and Improving Diagnosis

This thesis addresses critical challenges in healthcare through two distinct yet interrelated studies. The first paper tackles the escalating issue of financial fraud detection within healthcare systems, a pressing concern exacerbated by advancements in electronic payment methods. Conventional methods of fraud detection have proven insufficient; therefore, new ideas have to be developed. This work presents a comprehensive fraud detection classifier using machine learning techniques combined to improve performance. Methodologically, the ensemble classifier shows better than traditional approaches including Naive Bayes, Random Forest, as well as K-Nearest Neighbours by means of accuracy, precision, and recall measures. Having an accuracy of 99.46%, precision of 98.38%, and recall of 98.58%, the ensemble approach greatly beats its counterparts and presents interesting directions for next investigation. Additional research seeks to combine hybrid models specifically addressing dataset imbalances and guaranteeing real-time responsiveness in banking transactions. A crucial component of medical diagnostics with great consequences for patient care, the second paper tackles the urgent demand for fast and accurate detection of pneumonia from chest imaging (CXR) images. Using a novel deep learning architecture, the Swin Transformer V2, this work investigates its usage in the medical imaging domain for pneumonia diagnosis. Methodologically, the study assesses the performance of the model against a varied CXR dataset including several conditions and forms of pneumonia. Comparative analysis using well-known deep learning architectures including AlexNet, MobileNetV3, VGG-16, ResNet 50, and DenseNet highlights the accuracy of 98.6% by which the Swin Transformer V2 detects subtle patterns suggestive of pneumonia. The results highlight the transforming power of including advanced deep learning algorithms into clinical evaluation procedures since they provide previously unusual accuracy and clear the path for major changes in medical practices. This study has possible uses including the integration of innovative diagnostic models with clinical environments, so transforming healthcare methods. Future directions of study could be investigating hybrid models integrating deep learning with conventional diagnostic techniques and optimizing models for real-time implementation in clinical environments.