A New Architecture for Diabetes Prediction Using Data Mining, Deep Learning, and Ensemble Algorithms

It is a big challenge to diagnose diabetes in an early stage. This causes a health problem because it is a severe cause of death if it is not treated early or it can trigger many secondary diseases that impact the well-being of the patient. In this document, we present a new method to accurately predict this disease using data mining, deep learning, and ensemble algorithms. Data mining includes the processes of data preprocessing to make it more comprehensible and gaining insights from the dataset. This architecture is divided in 7 steps: First, the dataset is loaded. Second, the variables are analyzed to understand their value to predict diabetes. Third, the noise is removed from the dataset, deleting empty data. Fourth, the variables are transformed and scaled. Fifth, an exploratory analysis is made to explore the correlations between the variables. Sixth, the following predictive methods are applied: random forest, artificial neural network, and AdaBoost. Finally, results are presented and explained. To implement this method, we used a public dataset from kaggle called: diabetes dataset. This method achieved great accuracy, precision, and recall, which helps demonstrate the effectiveness of the method. Finally, this document could be the base for new research in this disease like trying to predict the type of diabetes the patient has, and it can be applied to different health problems. Furthermore, more predictive methods should be applied to try to achieve a higher accuracy.