Predictive Analytics: The Application of J48 Algorithm on Grocery Data to Predict Obesity

摘要

Current conventional methods of predicting obesity are costly where the government collects huge amounts of public health data through National Health and Morbidity Survey (NHMS) to do health forecast especially for noncommunicable diseases (NCD) including obesity. However, the advancement of technology can provide cheaper and easier obesity prediction methods. Instead of focusing on the traditional method of inspecting nutrition data from numerous populations based surveys, this study views from the perspective of `available' nutrition datasets which can be potential proxy to predict obesity prevalence. By applying grocery data and data analytics, various patterns of nutrition intake can be explored to predict the percentage of individuals in a household getting obese; this include calories, macronutrients, food groups, and food categories (raw & processed) food intake. Therefore, this study aims to test the potential of using grocery datasets to predict obesity. An experiment based on predictive analytics using J48 algorithm has been setup to measure the accuracy rate using the WEKA data mining tool. From the experimental result, the J48 algorithm predicted the obesity from the individual calorie intake patterns up to 89.41% accuracy, which is significant to represent nutrition data, and therefore implies that grocery data can be potentially used as an additional variable to the obesity prevalence trends in projecting future obesity burden.