Although many people are aware of the current climate crisis, most people do not actively pursue an environmentally friendly lifestyle. We were inspired to create a tool that could predict a user's level of environmental engagement and receptiveness to learning about living greener based on their current lifestyle choices. We believe such a tool can help drive impactful change as educators and environmental advocates can optimize their resource allocation to mobilize initiatives geared toward audiences that will be most receptive and responsive to this mission.
What it does
A neural network model was trained based on survey responses from college students indicating their lifestyle choices and knowledge/opinions about their carbon footprint. The created predictive model takes in a response from a user and generates a prediction about the user's sentiments toward living greener. The algorithm determines whether this person would be receptive to more information and if they would be likely to change their daily habits based on this, along with how much they care about their personal carbon footprint. Based on this, the application adapts to the user's current environmental engagement and responds with diverse strategies to persuade, educate, and raise awareness effectively.
How we built it
The neural network model was created using Keras and Python. The dataset used to train the model was downloaded from Kaggle. The interactive user-interface was built using QtDesigner.
Challenges we ran into
All categorical data from the dataset needed to be preprocessed to return numerical representations in order to be used in training the neural network model. This analysis proved to be somewhat complex. Tuning the model to achieve optimal levels of accuracy in prediction on sets of testing data and integrating the program with the user interface proved to be challenging as well.
Accomplishments that we're proud of
We are proud that the model ultimately achieved around 85% accuracy in predicting the user's sentiments and receptiveness about environmental impact. We are also proud that we were able to build this working system in a little over 24 hours!
What we learned
Team members all had very different skill sets coming into this project, so we were all able to learn some new programming skills from each other! Some of us learned about creating user interfaces, others were introduced to machine learning techniques for the first time!
What's next for AI-Powered Sentiment Analysis Tool for Sustainability
We are eager to further refine the various persuasive strategies presented to the user as the end of the program. We feel that this part would benefit from speaking more with domain experts who are actively involved in environmental advocacy work. With the collection of more data, the predictive model may also be improved for even higher accuracy than it currently exhibits.