Processing of missing data by neural networks

TL;DR

This paper introduces a theoretically justified neural network method for handling missing data by replacing neuron responses with their expected values, improving performance over traditional imputation without needing complete data for training.

Contribution

It presents a novel, general approach for processing missing data in neural networks that is easy to implement and does not require complete datasets for training.

Findings

Outperforms traditional imputation strategies

Applicable to various neural network architectures

Does not require complete data for training

Abstract

We propose a general, theoretically justified mechanism for processing missing data by neural networks. Our idea is to replace typical neuron's response in the first hidden layer by its expected value. This approach can be applied for various types of networks at minimal cost in their modification. Moreover, in contrast to recent approaches, it does not require complete data for training. Experimental results performed on different types of architectures show that our method gives better results than typical imputation strategies and other methods dedicated for incomplete data.