Accurate Pharmaceutical Data

Case study:

Forecasting Medical Prescription Volumes

Delivered with:
Find out how

The ask

A global pharmaceutical corporation was looking for ways to anticipate the number of prescriptions that would need to be issued each month. Their ultimate goal was to have an accurate quarterly and yearly forecast of their prescription volumes in order to make data-driven business decisions.


What We Did

To simplify this process, we divided it into two stages. First, we used a machine learning solution to identify whether or not a patient would need a prescription. Secondly, if the output from the previous step was positive (meaning that the patient would be issued a prescription), we used another machine learning solution to detect whether this would happen within the next 30 days or later. 

The Process

We used data preparation in order to work with different products in the pharmaceutical company. This data was also standardized so that it could be reached by different data sources in order to form a single dataset. 

An initial problem was encountered early into the process. After the data was analysed, it showed a very unbalanced result. It showed that there were a lot more cases of prescriptions to be issued as opposed to no prescriptions at all. To fix this issue, we applied oversampling and undersampling techniques driven by cross-validation. 

The most accurate information for the models is sourced from recent events. So, we automated the whole process to allow the models to re-train at a certain frequency, ensuring that we had the most accurate model ready before running the next predictions.



Our software was able to accurately predict if a prescription would occur within the next 30 days or more 90% of the time. This helped the company to make more informed decisions as well as provide better quarterly and yearly financial predictions. 

In our day-to-day business activities, many unbalanced results might happen. For example, these might include fraud detection, anomaly detection, churn prediction, and more. In these cases, applying the techniques we used in this case could make these results more accurate. 



For a free assessment, answers to your questions, or to discover how software and AI can be integrated into your business, don’t hesitate to get in touch.