Medical prediction from missing data with max-minus negative regularized dropout.

Area of Science:

• Medical research
• Machine learning
• Data science

Background:

• Missing data is a common challenge in medical research, impacting model generalization.
• Imputation techniques can lead to overfitting due to inherent uncertainty.
• Existing R-Drop methods struggle to differentiate sample types, hindering robust representation learning.

Purpose of the Study:

• To propose a novel negative regularization enhanced R-Drop scheme.
• To improve performance and generalization ability in medical prediction models with missing data.
• To address the limitations of standard R-Drop in differentiating positive and negative samples.

Main Methods:

• Developed a negative regularization enhanced R-Drop scheme.
• Introduced a max-minus negative sampling technique for diverse negative samples.
• Tested the method on three real-world medical prediction datasets with missing and complete data.

Main Results:

• The proposed method effectively boosts performance and generalization.
• Negative regularization enhanced R-Drop improves the learning of robust representations.
• Max-minus negative sampling provides sufficient diversity for the model.

Conclusions:

• The novel R-Drop scheme significantly enhances medical prediction models, especially with missing data.
• The method offers a promising approach for improving deep neural network generalization in healthcare.
• Effective handling of missing data and sample differentiation leads to more robust models.