Abstract — Multi-level disease prediction plays a vital role in the drug to disease discovery process. Most of the conventional models use static parameters and filtering approaches in order to filter the high dimensional feature space due to high computational time and memory. Also, these models are having less accuracy and high error rate for the classification models. In order to overcome these issues, a hybrid filtering method is proposed in order to optimize the data preprocessing and feature extraction on the high dimensional dataset. Experimental results proved that the hybrid data filtering and feature extraction models have better efficiency in terms of classification accuracy and runtime(ms) than the conventional models.