Article Preview
Top1. Introduction
Diabetes is a lifestyle disease with no cure. Its lifelong existence in the body gradually initiates other diseases and decays different organs. Most of the times diabetes remains hidden in the patient's body and do not show any syndrome. Currently, Diabetes Mellitus is spreading at an alarming rate. According to WHO (2016), about 422 million people were living with diabetes in 2014 and this number was estimated to increase to about 693 million by the year 2045 (Arnhold et al., 2014; Cho et al., 2018). About half of all people (49.7%) living with diabetes are undiagnosed, and the estimated healthcare cost of diabetic patients was USD 850 billion in 2017 (Cho et al., 2018). The WHO report also highlighted that 3.7 million deaths have been caused by diabetes (WHO, 2016). This alarming growth rate of diabetes is putting peoples' lives at risk worldwide, which is why it has become one of the foremost health concerns. Again, though many people have type 2 diabetes but still, its existence is not evident to them (Jourdan, 2012). Diabetes can be diagnosed through various types of blood tests which do not come handy and neither are they cheap. So, the rate of unawareness remains high. As diabetes is a hidden epidemic and a global health issue, predicting diabetes at an early stage or its probability beforehand gives the patients scopes to rebuild their lifestyle and food routine to save their lives. Thus, the development of an intelligent system for predicting the possibility of being diabetic becomes essential for the general people.
In the field of machine learning, the family of Naïve Bayes classifiers is regarded as one of the most common ways of predictive categorization using the probabilistic assumption of independent features (Wu et al., 2008). This group of algorithms has found their way into various fields like text categorization and analyzing documents (Chen et al., 2008; Schneider, 2005; Kibriya et al., 2004). In this research, Naïve Bayes classifier has been chosen primarily to propose an intelligent system for diabetic prediction despite the availability of other algorithms due to some specific reasons that includes: firstly, Naïve Bayes is remarkably simple to implement, has low computational complexity (Elkan et al., 1997) and provides very good accuracy (Ting et al., 2011). Secondly, Naïve Bayes classifier regards all the features equally for prediction. Finally, Naïve Bayes classifier is well known for its wide range application in healthcare prediction systems (Langarizadeh et al., 2016; Bhuvaneswari et al., 2012). Considering all these conveniences, Naïve Bayes classifier was considered as more prominent for implementing the diabetes predicting system compared to other machine learning algorithms.
Therefore, the objective of this research is to propose a Naïve Bayes based intelligent system for predicting diabetes. It is also worth mentioning here that an earlier version of this research was published in (Khan et al., 2017), where a mobile application was built for predicting diabetes based on the existing Naïve Bayes algorithm. In this research, an advanced algorithm is proposed to predict the possibility of being diabetic or non-diabetic more accurately and efficiently.