Regularized extreme learning machine for regression with missing data

This paper proposes a method which is the advanced modification of the original Extreme Learning Machine with a new tool to solve the missing data problem. It uses a cascade of L1 penalty (LARS) and L2 penalty (Tikhonov regularization) on ELM to regularize the matrix computations and hence make the MSE computation more reliable, and on the other hand, it estimates the expected pairwise distances directly on incomplete data so that it offers the ELM a solution to solve the missing data issues. According to the experiments on 9 data sets, the method shows its significant advantages: fast computational speed, no parameter need to be tuned and it appears more stable and reliable generalization performance by the two penalties, Moreover, it completes ELM with a new tool to solve missing data problem even when half of the training data are missing as the extreme case.