Prediction Pal

Inspiration

The inspiration was drawn from prior experience working with no/low code applications and seeing the value they added to businesses and individuals, Combining this with our background in Data Science and Computer Science, as well as the emergence of advanced machine learning. We wanted to empower the everyday user with the tools to perform advanced predictive analysis and machine learning.

What it does

Prediction Pal allows users to build and use machine learning models on their local device to generate predictions on unseen data. By uploading data and selecting a number of parameters users can take advantage of a variety of models to best suit their needs. Prediction Pal will then take the model the user has trained along with their new unseen data, to make predictions.

How we built it

The work was divided between the members as follows:

Rani: Front End (CSS & HTML)
Louise: Front End and Back End Integration
Sam: Backend and Machine Learning Integration
Gawri: Machine Learning Models

Challenges we ran into

We ran into a number of challenges, the greatest being trying to deal with Asynchronous functions, allowing the loading page to display whilst a model was still training or predicting.

Accomplishments that we're proud of

Being able to utilise async functions as well as being able to build the application within 2 days, using a framework (FastAPI) that no one in the team was familiar with.

What we learned

We learned a lot about async functions, managing version control in a team of 4 and using FastAPI.

What's next for Prediction Pal

We genuinely believe that Prediction Pal is commercially viable and will look to continue development. In addition to improving the UI and UX to create a smoother experience (such as uploading files rather than providing paths). We will also develop the following features:

Support for additional models (ARIMAX, Neural Networks and deep learning, XGBoost Regressor and more)
Support for Classification, Recommendation and NLP Machine learning systems.
Allow for hyperparameter tuning and optimisation for each model (e.g. Grid Search, Optuna, etc.) along with comprehensive documentation to help the everyday user.
The Ability to run multiple models concurrently
A dashboard to track the progress of models that are training/predicting.
An in app interface to allow users to view their data, select their target column and to feature engineer with Excel/Airtable like formulas.
Look at performance optimisation to improve run time of model training/prediction/optimisation
Support for GPU/TPU Processing
Ability to link with a mobile/tablet to provide push notifications on model completion (prior to this utilising SMS notifications through Twilio or email notifications through Microsofts Graph API.