Drug Prediction Based on Customer Reviews

All humans are prone to experience sickness at least once in their lifetime. Regardless of how sick humans get, there is a need to take medication most of the time. Often, doctors choose the appropriate medication from available drugs depending on the patient’s situation and the side effects of the medication. Given that we find 215,063 data points in which customers rate their satisfaction with various medications, we create two different models and use various machine learning techniques to determine drugs with fewer side effects and higher satisfaction rates among consumers. We want this method to generate a smoother relationship between physicians, patients, and medication options. We do not intend to generate a model that recommends medication to patients. Instead, we hope to create a tool that doctors can use to choose the best available option for their patients based on other patients’ previous experiences with the particular medication. This paper examines different preprocessing techniques, such as sentiment analysis and feature encoding, to clean and prepare data for applying different machine learning techniques. In addition, more than eight distinct machine learning techniques are used to train and test our data. In this paper, two different groups of conditions are used. One group focuses on a single condition, and the other examines the nine most distributed conditions. Then, we compare two models and realize that the model, which is focused on a single condition, gives us more accurate results than the nine most distributed conditions.