Implementation of Random Forest in Classification Model of Diabetes Prediction based on Drug Review Content

Diabetes is a disease that causes abnormally high glucose levels in the blood and is considered the main factor of damage to blood vessels. This disease can affect the heart, eyes, kidneys, and nerves and cause various complications. The risks of diabetes can be reduced with proper treatment in the early stage. Early identification of diabetes can be made by implementing a machine learning model to predict or diagnose diabetes. In this study, we utilized Random Forest (RF) to develop a classification model to predict people with diabetes. Random Forest is known as an ensemble method that combines the prediction of each basic model. The model development was performed by using drug reviews content retrieved from the UCI Machine Learning repository. The data set consists of the review and confirmed patient condition. We performed preprocessing step to obtain the important feature from the text. Then the feature extraction was done by using N-gram and term frequency-inverse document frequency (TF-IDF). We utilized six N -gram methods, i.e., unigram, bigram, trigram, unigram - bigram, bigram - trigram, and unigram - big ram - trigram. The best result of the experiments is obtained from the model developed by using a unigram feature with F -1 score is 0.952.