Development of machine learning model for diagnostic disease prediction based on laboratory tests

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Computational Biology

Background:

• Machine learning (ML) and deep learning (DL) are increasingly utilized in medical science.
• Applications span visual, audio, and language data processing for healthcare.

Purpose of the Study:

• To develop an optimized ensemble model for disease prediction using laboratory test results.
• The model blends a deep neural network (DNN) with two ML models.

Main Methods:

• Selected 86 laboratory test attributes based on clinical importance and data quality.
• Constructed LightGBM, XGBoost, and DNN models using TensorFlow on a dataset of 5145 cases (326,686 results).
• Investigated 39 diseases based on ICD-10 codes.

Main Results:

• The optimized ensemble model achieved an 81% F1-score and 92% prediction accuracy for the five most common diseases.
• Analyzed predictive power differences and classification patterns between DL and ML models.
• Utilized confusion matrix and SHAP values for feature importance analysis.

Conclusions:

• The developed ML model demonstrates high efficiency in disease prediction and classification.
• This approach offers valuable support for disease prediction and diagnosis.
• The study highlights the potential of ensemble models in medical data analysis.