Ou Deng, Shoji Nishimura, Atsushi Ogihara, Qun Jin
The Statistical-Neural Interaction (SNI) framework offers an interpretable method for mixed-type data imputation by integrating statistical priors with neural attention, balancing accuracy and interpretability.
This study introduces a new approach for filling in missing data in tables that include both numbers and categories, which is a common problem in many real-world databases. The method, called Statistical-Neural Interaction (SNI), combines traditional statistical methods with modern neural network techniques to predict missing values. Unlike some other methods, it not only predicts missing data but also provides insights into which variables were most influential in the predictions. While it may not always be the most accurate for categorical data, it offers valuable interpretability, making it useful in scenarios where understanding the decision-making process is important.
