Improving Health-Care Systems by Disease Prediction

Health-care generates a large amount of data, the task is to collect this data and effectively use it for analysis, prediction, and treatment. A better approach to health-care is to prevent a disease with early intervention rather than taking a treatment after it is diagnosed. There are systems that give information on treatment for diseases but not a lot of systems predict the possibility of having a disease. The aim of this research is to predict whether a person might have diabetes or not. The data set will contain the patient ID, gender, age, symptoms, glucose level, blood pressure, BMI etc. The outcome will be either 1 or 0, where, 1 meaning the patient is diabetic and 0 meaning the patient is not. There are 4 types of diabetes: Gestational Diabetes, Pre-diabetes, Type 1 Diabetes, Type 2 Diabetes. The system will predict the type 2 diabetes the person might have based on the data provided. The main goal of the project is to try to increase the accuracy of the prediction. This can be done using machine learning prediction algorithms. It works well in handling multidimensional and multi-variety data in dynamic environments. One of the limitations is machine learning needs a lot of training data which might be cumbersome to work with. The outcome of this research depends on how accurate the data set is. This paper also focuses on a comparison of 5 prediction algorithms which are Artificial Neural Networks, Logistic Regression, KNearest Neighbors, Decision Tree and Random Forest algorithm. According to the results, neural networks produce the highest accuracy after model evaluation.