Real-Time Predictive Analytics for Prescription Deliveries

Tech Stack

Microsoft Fabric, Power BI, EventHouse, One Lake, Dataflow, Azure ML, Azure SQl Storage, Python, MLFlow, PySpark, Scikit-Learn

Problem

The client wanted to figure out whether a prescription delivery would be late or not, and if it was late what were the reasons behind it being late. The client requested us to make it as easily understandable as possible.

Solution

1. We created a solution for the client fully on Microsoft Fabric using services like Lakehouses, Dataflows, Notebooks and Reports. The solution used advanced ML models made using MLflow experiments to predict the probability of the delivery being late.
2. We also implemented model explanation algorithms to let us know the reasons behind a specific prescription delivery being late according to the ML model. It lets us interpret the ML model’s output in a clear way.
3. We created an Active Learning pipeline within Fabric to continuously train the model as new data flowed in from the client. This ensured that the model would improve and learn more as time passes.
4. We also utilized PySpark to create distributed computing clusters on Fabric to run our ML model experiments, model explanation algorithms and inference & training scripts at high speeds.
5. We also created a Power BI report to visualize the results from the ML model, monitor the accuracy of the model and summarize historical data in a visually appealing manner.
6. We utilized Dataflow Gen 2 pipelines to transform our data at various stages of the solution, like fetching from Azure SQL Storage and feeding the data to the ML model.
7. The solution ran in real-time using Eventhouse in Fabric letting the client know the latest updates as soon as possible.

Impact

Using the solution the client was able to figure out the main reasons behind the prescription deliveries being late in the past as well as the future.

The visually appealing Power BI report let the client be able to learn about the most important decision factors and look at the changes in them with each delivery.

It enabled him to expand the business, enhance operational effectiveness, and achieve high levels of customer satisfaction.