Impact of Data Preprocessing and Quality Assurance on the Prediction of Recurrent Neural Network-Based Clinical Deterioration

This is the authors' abstract. We don't add key points for this paper.

Early detection of patient deterioration is crucial for improving patient outcomes. Machine learning (ML) techniques have shown considerable potential for supporting clinical decision-making; however, their predictive performance depends strongly on the quality of healthcare data. This study investigates the influence of patient data quality on the development of patient deterioration prediction models by proposing a recurrent neural network (RNN) framework using routinely monitored physiological measurements from the Cleveland Heart Disease database. The proposed framework incorporates comprehensive data preprocessing, including data cleaning, normalization, and exploratory data analysis, to improve data reliability before model development and Modified Early Warning Score (MLEWS) calculation. The RNN model was trained, tested, and optimized, with its performance evaluated using both classification and regression metrics. Experimental results demonstrated excellent predictive performance, with classification accuracy, precision, recall, specificity, and F1-score ranging from 0.992 to 1.000, while an AUC value ranged from 0.992 to 1.00. Regression evaluation also indicated negligible prediction error, with a mean absolute error (MAE) and mean squared error (MSE) of 0. Cross-validation further demonstrated the robustness and generalizability of the model, with validation accuracy ranging from 94.6% to 100%. These findings indicate that comprehensive data preprocessing and quality assessment can substantially improve the reliability and predictive performance of RNN-based clinical decision-support systems. The proposed framework provides an effective approach for supporting the early detection of patient deterioration and facilitating timely clinical decision-making during emergencies