In healthcare informatics, the individual’s disease prediction and its classification are essential. With the utilization of data mining systems, we can analyze the disease at a beginning stage and enhances the patient's survival rate. But still, it has some issues like removing missing values and feature selection from the medical datasets. To overcome that, optimal features are selected from the datasets by the use of an innovative optimization algorithm. In the proposed work, multi-datasets (Liver, Lung, Heart, and Thyroid) are considered for the disease prediction analysis. Initially, the missing values from the input datasets are removed during the preprocessing stage. Next, to that, the optimal features are chosen by the algorithm called Modified Monarch Butterfly Optimization (MMBO). Based on the selected features, the data are classified into two norms: healthy and non-healthy by the proposed classifier i.e. Deep Neural Network (DNN). The performances of the proposed algorithm and classifier are tested on the multi-datasets in terms of sensitivity, specificity, and accuracy. The results demonstrate the MMBO-DNN algorithm achieves high accuracy of and less execution time compared to existing algorithms. © BEIESP.