In recent years, the rapid surge of electronic medical record (EMR) data has underscored the need for efficient and precise disease classification models. Precise disease classification can significantly impact clinical decision-making, enhancing patient outcomes and resource allocation. However, existing models for disease classification in medical records have demonstrated limitations in their accuracy, recall, precision, and timeliness of classification, leaving room for potential misdiagnoses and delayed interventions. To address these gaps, this paper introduces a novel fusion of BiLSTM-BiGRU architectures with Autoencoders for advanced feature representation, followed by the application of Deep Q-Networks (DQN) for the classification of clinical events in medical records. Our proposed model, when benchmarked against prevailing methodologies, showcases marked improvements. Specifically, it enhances the precision of disease classification by 1.5%, accuracy by 2.9%, recall by 4.5%, and AUC by 3.9%. Importantly, the proposed framework also reduces the delay in disease classification by a significant 10.4%. The enhanced performance metrics underline the potential of our model in various clinical scenarios, ranging from early disease detection to improved patient management. The integration of this approach into healthcare systems can drive timely and accurate clinical interventions, furthering the broader goal of personalized patient care and optimized healthcare delivery.