TouchGrass

A bar chart charting the importance of each variable in calculating the sustainability score of individual data points during training.
A classification report for the categorization of flowers, showing the precision, recall, and F1 score for each of the 12 flowers.
A heat map to show the correlation between variables used in the predicting of sustainability score.
Output of the flower recognition model
Challenge flower photo upload
Problem description
Flower Classification accuracy chart
Sustainability output score
Survey questions for sustainability
touchGrass landing page with logo

The Problem

Spending prolonged periods indoors, particularly for STEM majors who often engage in intense study and screen time, can indeed negatively impact mental health. A study found that college students who spent more time on screens (5+ hours daily) had a 48% higher risk of depression and anxiety compared to those with lower screen time.

Research also shows that individuals who spend less time outdoors are 20% more likely to experience symptoms of anxiety and 28% more likely to report symptoms of depression.

We also noticed that a lot of the marketing for having a more sustainable lifestyle is impersonal and the advice isn't very specific and doesn't encourage mindfulness as a lifestyle.

We wanted to see if we could tackle both highly integrated problems at once in a way that makes it personal to the user.

The Solution

touchGrass: a website that uses humor and socialization to promote going outside and uses that experience to encourage thoughts about sustainability as a lifestyle. It just takes six simple steps to use touchGrass.

Get sent the link.
Go outside.
Take a picture of a flower.
Learn about the flower, its connections to mental health, and how it is endangered by climate change.
Find out how sustainable your lifestyle is.
Reflect, consider, and keep promoting mental health by sharing the link.

How it works

User Interaction: The application has a user-friendly interface where users can upload an image of a flower and fill out a form to assess their sustainability score. Uploaded images are processed, analyzed, and classified into 14 specific flower categories: carnation, iris, bluebells, golden english, roses, fallen nephews, tulips, marigolds, dandelions, chrysanthemums, black-eyed daisies, water lilies, sunflowers, and daisies. Form submissions are used to generate a sustainability index score.

Image Classification: When a user uploads a flower image, the app preprocesses it using MobileNetV3, a neural network model. It then classifies the image, assigning it a category (e.g., Rose, Tulip, Dandelion). The classification results provide users with specific details about the flower’s symbolic meaning and how climate change affects it.

Sustainability Index Prediction: The sustainability score prediction uses a Random Forest model trained on lifestyle-related data, like diet, transportation, and energy sources. Users input details about their lifestyle choices, and the model calculates a “sustainability score” based on encoded features. The score is then displayed, helping users understand their environmental impact.

Data Processing and Model Integration: For the sustainability score, categorical features are encoded using LabelEncoder based on predefined mappings. The LabelEncoder transformations ensure compatibility between user input and the machine learning model, providing an accurate score.

How we built it

We used Flask for the back-end web server, which connects python code with HTML, CSS, and JavaScript templates. Flask’s simplicity helped structure the project into clear sections, like image classification, file upload, and form submission.

For flower classification, we fine-tuned MobileNetV3 on a dataset of flower images, achieving a high accuracy of 97% through techniques like data augmentation and transfer learning. For the sustainability index, we trained a Random Forest model on a structured .csv dataset that included lifestyle attributes that were easily known to the user. The model leverages encoded categorical data to predict sustainability impact, with a val accuracy of 76% because of eliminating variables that are hard for the average college student to input into the website. Data: Sustainability Dataset Flower Dataset