Predicting medical drug usage intentions via SGD-based text classification model

The effects of medical drugs and their usage purposes vary among individuals due to the chemical composition of drugs, side effects, genetics, etc. Even if those effects are to be discovered pharmacologically, they cannot be fully understood. Hence, it becomes essential to analyze the individuals’ reviews and experiences to unearth such effects and find out which other purposes drugs are used for, in addition to the target disease they are developed to cure. Text classification methods present various solutions to analyze those reviews effectively. Generally, these effects are investigated in terms of emotional analysis of medical drug usage experience as positive or negative. However, some drugs can be used for more than one specific treatment. For example, an antipsychotic drug can be used for both depression and anxiety or ADHD. Therefore, the effects of medical drug users and drug names to be associated with the review of the studies should be covered comprehensively. Based on this motivation, this study proposed a lightweight model for the prediction of medical drug usage intentions using text-based patient reviews. For this purpose, TF-IDF and bigram methods are used for text classification in the feature extraction step, then the Stochastic Gradient Descent (SGD) classifier is used for prediction and compared to other popular machine learning algorithms. Classification results indicate that the SGD and TF-IDF-Bigram approach effectively predicts drug usage intentions for medical purposes with an accuracy of 98.42%. Based on the outcomes, it is concluded that the findings of this study may be beneficial in pharmaceutics or medicine considering drug design, reducing side effects, health management, treatment adherence and process design, and personalized medicine.