Democratizing Study Space with
Predictive Forecast Machine Learning
Predictive Forecast Machine Learning
StillSpace is an Android application which shows study space hotspots and current usage, crowdsourced from students who recently studied in the area and processed with machine learning.
The project was created during September 6-8, 2019 for PennApps XX Fall 2019, at the University of Pennsylvania - a 36 hour hackathon, the 20th iteration of the first-ever student hackathon, and one of the largest in the world with 1200 attendees.
Students today are under immense pressure to do well in difficult academic situations and achieve high grades to be accepted into exclusive institutions. In order to support this, universities often design specifically-designated study areas for students to use.
Though the spaces are often excellent for studying, supply rarely if ever satisfies demand. Expensive renovations cannot meet the needs of smaller universities, let alone for universities with over 20,000 enrolled. The result is that students consistently have a significant lack of study area with enough space and quiet in which they can think and learn at their best capacities. Students often must make do with loud, crowded areas, spend time travelling to a different location off-campus, or skip studying due to the demoralizing hunt for suitable study space.
We isolated our key target audiences to be university students from high school, continuing learners, and self-educators. These people have enough time or need to warrant heavy studying to learn sufficiently to do well, but are often deprived of the ability.
Our solution is StillSpace, an Android application which displays study space hotspots. These hotspots display relative availability of study areas and may also reveal previously unknown study spots. Over time, a population using StillSpace will redistribute students to better and more fairly utiilize all possible study areas.
Data for StillSpace is crowdsourced from other students who recently studied in the area and used the application. It is processed with machine learning algorithms from AWS Forecast to provide predictive forecast data for the future.
The mobile application on Android was the user-facing section of StillSpace, providing a locator and predictor for study space availability as well as tracking location data during studying.
Download the application on the Google Play Store here.
The application utilizes Google Location Services as well as Google Maps Heatmap Utility, an advanced and less-known Play Services API.
The map view is set to automatically display Towne Building in the Engineering Division at the University of Pennsylvania, as this was the location of the hackathon and demo.
To understand how users would experience the application, interface mockups were created using Figma. Once complete, assets were then passed to the developers with Zeplin.
The machine learning component of StillSpace utilizes Amazon Web Services Forecast, an online machine learning service which processes trends in time series data to predict future forecasts.
Using Amazon Forecast also entailed creating an Amazon S3 bucket which contained a CSV file to be imported.
We encountered heavy limitations when working with the machine learning framework. First, a sufficient amount of suitable sample data which mimicked actual conditions was incredibly difficult to generate. Second, there was not nearly enough time and data to train the network sufficiently well in a single weekend. Third, we initially hoped on having the data be automatically uploaded to the server and to configure the machine learning to be run on the dataset on a set recurring schedule. Unfortunately, the Amazon Forecast API is still greatly under development, and did not have this capability. Instead, each dataset needed to be run automatically using the web console.
The services available for machine learning are still in the process of being developed and made ready for public use but, when ready, they will empower even students to easily be able to find valuable trends across vast datasets.
Watch a video of our pitch here:
Disclaimer: This page was created after the hackathon's completion.