Extreme Learning Machines (ELM) is a new learning algorithm for Single hidden Layer Feed-forward Networks (SLFNs). The ELM has better generalization, rapid training and lower complexity, however, the method suffer from singularity problem and obtaining optimum number of neurons in the hidden layer. In this paper, we considered an IHT for sparse approximation of the output weights vector of the ELM network. The performance evaluation of the proposed method which is called IHT-ELM, was chosen out on four commonly used medical dataset for prediction purposes. The results showed that IHT-ELM has several advantages against the original ELM methods such as obtaining optimum number of neurons and low complexity.