Auto Assignment of Transport Requests for Acuity Link 

The Challenge

The platform has a set of transport providers who service requests from healthcare institutions. These requests, sent to a set of transport providers, are needed to be accurately assessed for acceptance or rejection. This manual process is time-consuming and often leads to delays in patient transportation. The client sought a data-driven solution to automate and optimize this process. 

The Solution

Based on historical data, there is always a higher probability of one transport provider being able to service certain types of requests. The challenge was to identify the transport provider and assign the requests to them on priority, thereby saving precious time and effort. The overall solution approach and steps are detailed below:

Data Collection and Preparation:

• Extract historical data on transport requests, including date & time, requesting facility, pick-up and drop-off locations, request status, and service type.
• De-identify data where necessary to protect PHI. 

Machine Learning Model Development:

• Build a Machine Learning (ML) model to predict the likelihood of a transport request being accepted by the provider.
• Utilize features such as requesting facility, service type, pick-up location, and drop-off location for prediction.
• Train the model on historical data to make accurate acceptance/rejection predictions.

Model Deployment:

• Since this is a classification problem, we decided to begin with Naive Bayes but also use k-Nearest Neighbours and Random Forest to further validate our understanding
• Deploy the trained ML model within the healthcare facility’s system.
• Integrate the model into the transport request workflow for real-time decision-making.

Continuous Improvement:

• Implement a system for ongoing model re-training using new transport request data.
• Monitor model performance and accuracy.
• Make necessary updates and improvements to the model to enhance prediction capabilities.

The Outcome

By leveraging Machine Learning and data-driven decision-making, our client has successfully automated and optimized the transport request process. This has not only improved the efficiency of patient transportation but also enhanced the overall quality of healthcare services provided. Continuous monitoring and re-training of the ML model ensure that it remains responsive to changing conditions and continues to deliver accurate predictions.

Tech Stack:

Python
AI-ML Models: Naive Bayes, k-Nearest Neighbours and      Random Forest
DB: DynamoDB, Cassandra