Abstract:
This paper examines various machine learning methods for assessing risk
factors for cardiovascular diseases. To build predictive models, two
approaches were used: the extreme gradient boosting (XGBoost) algorithm
and a convolutional neural network (CNN). The focus is on analyzing the
performance of each model in classification and regression tasks, as well as
their ability to identify key biomarkers and risk factors such as cholesterol,
ferritin, homocysteine and aspartate aminotransferase (AST) levels.
XGBoost parameters have been optimized for working with tabular data,
demonstrating high accuracy in risk prediction. The CNN model, despite the
initial reduction in error on the training set, showed signs of overfitting when
analyzing validation data. Performance evaluation using the metrics of mean
squared error (MSE), coefficient of determination (R²), Akaike information
criterion (AIC), and Bayesian information criterion (BIC) revealed
significant differences between the models. The study results confirm the
effectiveness of XGBoost in analyzing tabular data and summarizing risk
factor knowledge, while the CNN model needs further optimization to
handle sparse data. The work demonstrates the importance of choosing the
right model architecture and training parameters to ensure reliable diagnosis
of cardiovascular diseases.
