Підвищення ефективності медичних систем підтримки прийняття рішень на основі машинного навчання та хмарних сервісів

The object of research is the process of developing and deploying decision support systems using ML models and cloud infrastructure. A significant problem solved in this research is the software implementation of such DSSs with ML models, as well as their further deployment for end-user access. As a result, a multipurpose scheme that combines the stages of local development and publication in a cloud infrastructure is proposed. Such approach is relevant for small companies and government agencies as it allows them to save financial resources on maintaining permanent IT specialists, maintenance and support. Its distinctive feature is that model training and its integration into a web application are performed at the local stage, while the publication stage uses cloud services to automatically update the project. The research implements a comprehensive data preprocessing pipeline for stroke risk prediction, including KNN-based imputation for missing values and SMOTE + NCL for class balancing. Following a correlation analysis and data augmentation four classification algorithms: logistic regression, SVM, Random Forest, and eXtreme Gradient Boosting were evaluated. Logistic regression is identified as the top-performing model regarding recall after data augmentation. The final model is integrated into a Flask application via serialization and a dedicated inference module. The application is published automatically from GitHub to Amazon’s cloud environment using such services as EC2, S3, ECR, and Secrets Manager. The cost of maintaining such a project is significantly lower than using dedicated servers or third-party software with a subscription fee per user. The results can be used in various industries to create DSSs that require high availability and minimal maintenance costs.