[Features Interaction Lasso for Liver Disease Classification].

Area of Science:

• Medical Informatics
• Machine Learning
• Biostatistics

Background:

• Hepatitis disease classification presents complex challenges due to intricate feature interactions.
• Existing classification methods may not fully capture these interactions, limiting accuracy.

Purpose of the Study:

• To propose a novel lasso method that integrates feature interactions for enhanced hepatitis disease classification.
• To evaluate the effectiveness and efficiency of the proposed method against established techniques.

Main Methods:

• A new interactive model was defined, incorporating a lasso penalized function and hierarchical convex constraint.
• The model was solved using a convex optimization approach, combining Karush-Kuhn-Tucker (KKT) conditions with generalized gradient descent.
• Sparse solutions for main effect and interactive features were derived to implement the classification model.

Main Results:

• Experiments on two liver datasets demonstrated that feature interactions significantly contribute to liver disease classification.
• The proposed feature interaction lasso method exhibited superior effectiveness and efficiency compared to standard lasso, all-pairwise lasso, SVM, KNN, and LDA.
• The method showed strong explanatory ability in identifying key features and their interactions.

Conclusions:

• Feature interactions are crucial for accurate hepatitis disease classification.
• The proposed feature interaction lasso method offers a powerful and interpretable approach for disease classification.
• This method outperforms traditional machine learning algorithms in classifying liver diseases.