Industry: Health & Fitness
Services: AI & ML, Web & Mobile development, Product Engineering
In the field of health and fitness, the Client offers user-friendly and personalized apps. Over 75 million people use the custom-created apps across platforms to improve their nutrition, sleep, physical exercise, and mental well-being.
The client strongly believed that personalized content is the game changer in health and fitness applications. Using the data from the customers and running the data through several models already in place can significantly generate personalized recommendations and solutions to cater to health and fitness requirements of the user.
The existing database design and data quality was inefficient for analytics. Metricoid team conducted relocation of entire data from the client’s database to inhouse storage and revamped the database structure of the client accordingly. Our next step was to work on data visualization by mapping out the tables of data. As a raw data storage, there was a Couchbase cluster. For a reliable data transfer to BI data warehouse (DWH) we added Kafka , i.e. PostgreSQL, using Spark procedures. To provide interactive dashboards, we used Apache Superset on top of BI DWH.
As per requirements of the client, a solution was needed wherein the data analysts and marketing managers the option to select from different filter parameters such as:
To enable a higher personalization level, several dashboards were included.
To allow the existing app to deliver personalized recommendations, a major challenge was to create a recommender system using Machine learning and predictive model. The ML based model was trained based on the data that users share through their profiles. A user enters basic personal data after downloading and installing the app, and the algorithm chooses a best-suited model and assigns a plan from the data provided.
Also written in Python is a tree-based recommender system and associative rules mining using the Apriori algorithm. Automatically generated recommendations vary based on a user’s fitness level. If a user’s personal data indicates that they are unable to cope with the present set of activities, the system can advise repeat workouts at the current level, or higher-level workouts.
The client provided all the data for training. One major source of data gathering was from GPS provided it is allowed by a user and other data were based on the user activity and inputs.
The Metricoid team has enhanced the existing mobile fitness apps of the Client with predictive analytics capabilities.