A computational approach of data smoothening and prediction of diabetes dataset

Data mining when applied on medical diagnosis can help doctors to take major decisions. Diabetes is a disease which has to be monitored by the patient so as not to cause severe damage to the body. Therefore to predict diabetes is an important task that is most important for the patient. In this study, a new data smoothening technique is proposed for noise removal from the data. It is very important for the user to have control over the smoothening of the data so that the information loss can be monitored. The proposed method allows the user to control the level of data smoothening by accepting the loss percentage on the individual data points. Allowable loss is calculated and a decision is made to smoothen the value or retain it to the level which is accurate. The proposed method will enable the user to get the output based on his requirements of preprocessing. The proposed algorithm will allow the user to interact with the data preprocessing system unlike the primitive algorithms. Different levels of smoothened output are obtained by different loss percentage. This preprocessed output produced will be of a better quality and will resemble more to the real world data. Furthermore, correlation and multiple regression is applied on the preprocessed diabetes dataset and a prediction is made on this basis.