The association of operating systems, and networks in healthcare environment increases risk of data security due to cyber-attacks. However, previous cyber-attack detection model fails to detect attacks in healthcare system accurately and needs more resources. To overcome this problem, a Modified Recurrent Neural Network (RNN) is employed for cyber-attack detection and to classify attacks as per labels. TON-Internet of Things (TON-IoT), Intensive Care Unit (ICU), Extensive Care Unit-Internet of Health Things (ECU-IoHT), Washington University in St. Louis Enhanced Healthcare Monitoring System (WUSTL-EHMS) datasets are preprocessed by one-hot encoding method and min-max normalization technique. After that, features are selected using XGBoost Regressor and irrelevant features are eliminated by Recursive Feature Elimination (RFE) algorithm. Finally, accurate detection of cyber-attacks is performed by proposed approach. The experimental analysis indicates promising results for detection of cyber-attacks in healthcare system with accuracy of 99.2%, recall of 99.1%, precision of 99.1%, and F1-score of 99% which is superior than existing detection methods like LightGBM and Transformer, Extreme Learning Machine (ELM) and Bayesian optimization, Ensemble Learning, Directed Acyclic Graph -Long Short-Term Memory (DAG-LSTM) and Quantum Deep Neural Network (QDNN).