Machine Learning Recommender Engine

Project Description:

Caremada provides an online service that connects caretakers with patients in need of specific or general care. Caremada’s service is similar to Uber in that service providers book their own time slots to work and clients find a suitable provider. Caremada has two core users; caregivers, and clients. Caremada categorizes its caregivers by the services they provide (caretypes). Carmada wished to implement machine learning tools to provide better assistance in finding caregivers and services to nearby patients.

Caremads is a startup company, meaning they have not gone to market yet, and Artificial Incoherence is called to implement a foundation to what will become a caregiver recommendation system used by their clients. With the current technology used, Caremada was struggling in connecting carertakes with patients. There was a lack of connection between the users, causing confusion. The requirement that the client needed was a proof of concept on a recommendation system, which they can later implement with their datasets.

This project was broken down into two phases. The first phase was the actual creation of a recommendation system. We first started a research phase, we went through different machine learning algorithms, to pick the best algorithm which would suit the clients needs. We ended with: Cosine Similarity, and Single Value Decomposition (SVD). This algorithm best works on “text base”, meaning we are able to make an accurate prediction based on the type of care the client chooses. The S.V.D algorithm makes predictions based on user-to-user interaction. This means, it makes predictions based on similar patterns of patients who have similar needs.
The second phase of the project was to implement a web interface to easily interact with the recommendation system, and to provide a visual outcome of the system . There is little use to have a recommendation system where the data cannot be easily seen by the user. We converted our recommendation system into a standalone server, where the web interface can communicate and retrieve the most suitable solution data for the user. We also implemented a database that would work hand in hand with our recommender system.

Throughout the progress of this project, this has provided us a chance to understand higher-level math that is involved, and new methodologies used by machine learning, as well as new software that is used in the industry. The final prototype was above and beyond the client’s expectations, as we were provided very little resources in data to begin with. We proved the data structure required for the startup company business model.

Contact the Team

Funded By