Inspiration
Through the years, we have met and interacted with many people who were bullied or left out for not being able to integrate well into social circles. It was not till many years later that they were diagnosed with Autism Spectrum Disorder (ASD), allowing them to better understand themselves and find supportive communities.
This challenged us to learn more about the issues that neurodivergent individuals faced and fueled our desire to improve the lives of those affected. With significant challenges in early detection and intervention, our team was inspired by the potential of technology to transform the diagnostic process. We wish to harness the power of machine learning and artificial intelligence to bridge the gap between the initial signs of autism and the tailored interventions that can make a world of difference for those on the spectrum. This project represents a commitment to fostering understanding, inclusivity, and improved quality of life for individuals with autism and their families.
What it does
We envision a mobile application that not only detects autism but also provides the service in an affordable and accessible manner. With a mobile device and internet connection, anyone will be able to take a short diagnostic test that uses eye movements and patterns to identify when an individual shows the typical signs on someone on the autism spectrum.
How we built it
We started off by analysing the types of data provided by the eye-tracking dataset and understanding what can be identified. This included the participants diagnosis status, relative eye position and blinking patterns. After cleaning the dataset to remove datapoints without useable information (e.g. unidentified participates), we sourced multiple training models and decided upon using tensor flow for its efficiency. The cleaned dataset is then trained and tested at a 7:3 split, which gives an accuracy of around 70%.
Challenges we ran into
We used a dataset from Kaggle that was slightly difficult to manipulate since there was some missing data and unidentified individuals. Hence, we had to do a few rounds of data cleansing to get our relevant data. Since the dataset only tested 59 patients, we also had a limited amount of data that could be used to train the machine, which was a possible limitation to its accuracy.
Accomplishments that we're proud of
We are glad to have produced a user-friendly interface for children to test for ASD with an accuracy of approximately 70%.
What we learned
It is essential to understand the data that we are have/need and the project outcome, so that we are able to appropriately manipulate the data and draw conclusions.
What's next for Eye-tracking to help diagnose of Autism Spectrum Disorder​
More data can be collected from a much larger group of children, and of different nationalities and backgrounds.
Log in or sign up for Devpost to join the conversation.