Data Science Showcase Winter 2020

Despite the COVID-19 outbreak, hundreds of BrainStation students recently graduated with Diplomas in UX Design, Web Development, and Data Science. We couldn’t be prouder of their achievements and the way they took everything in stride.

BrainStation’s full-time Diploma programs culminate with students producing an individual portfolio piece for their final Capstone project.

Here are some of our favorite Data Science projects from Winter 2020. You can also see all of our graduate’s projects on our hiring page.

RoadBoost: Intelligent Traffic Light Control System

Daria Aza

The satisfying video below speaks for itself. See how Daria was able to improve traffic flow using Deep reinforcement learning to regulate traffic lights. 

Using SUMO (Simulation of Urban Mobility) to create an environment with vehicles and pedestrians, Daria’s goal was to teach traffic lights to react to situations in real-time and adjust lights accordingly. The negative cumulative reward was calculated based on the number of vehicles and pedestrians stopped at the intersection. After every action, the reward was recalculated, eventually decreasing the delay to such an extent that most vehicles and pedestrians were able to move through without waiting.

View Daria’s GitHub to learn more. 

Human Activity Recognition

Carlo Jiaxu Chen

Carlo developed a novel attribute-based machine learning architecture that can recognize human activities such as walking, running, or laying down with limited training data, and an efficient neural network model (CNN-LSTM). The model can accurately predict your activity with results of up to 94 percent accuracy!

He intends for this technology to be useful for virtual reality software, health, and fitness services, and more.

See Carlo’s GitHub to see more of his process.

Classifying Letters of the ASL Alphabet

Angel Phanthanourak

Angel sought to classify letters of the American Sign Language (ASL) alphabet in real-time. Using deep learning (specifically the VGG16 Convolutional Neural Network) she was able to predict ASL letters from live webcam capture with 66 percent accuracy, much higher than randomized predictions, which are accurate 4 percent of the time.

Her hope is that this tool can help accessibility and independence for members of the deaf, hard of hearing, and non-verbal communities.

Find updates on Angel’s LinkedIn.

Forecasting Import and Export Trends

Reza Navidi

Reza’s time series model forecasts import and export trends in Canada. The model investigates global market shares and finds the top merchandise in import and export over the years.

See his work on GitHub.

Radiomics: Associating tumor features with survival outcome

(Hervé) Hiu Fai Choi

Hervé’s project aims to associate lung cancer tumor features with the survival outcome of the patient. Observing characteristics of the tumor, such as how big, irregular, lumpy, or coarse the tumor is, may influence the treatment the patient chooses to undergo or the clinician chooses to administer.

Herve’s data found that more non-uniform tumors yield unfavorable survival outcomes, which may prompt more aggressive approaches for tumor management.

Look into more of Herve’s work. 

Use of NLP and Supervised Learning to Target Wine Scores

Aviel Stern

Aviel created a predictive model for wine scores by looking at geography, price, description, variety, and vintage. This algorithm provides valuable consumer insights to wine sellers curating selections and consumers looking for trendy flavors.

Her work found that logistic regression performed the best with a 80 percent accuracy score in predicting how well wine would be rated.

See Aviel’s process on github.

NLP for Fake News Detection

Tom Satterthwaite

The public is only able to identify fake news headlines with 16 to 36 percent accuracy, so Tom aimed to use NLP on news article headlines to develop a model with 75 percent accuracy.

In the end, Tom was able to build an RNN to achieve an accuracy of 82 percent in predicting fake news.

Take a look at Tom’s GitHub.

Genre Genie – Movie Genre Predictions

Tom Keith

Tom’s app predicts movie genres from a plot summary using multi-label classification machine learning and natural language processing.

Just enter a plot summary and Genre Genie will predict the genre breakdown between themes such as action, fantasy, adventure, sci-fi, or drama with up to 100 percent accuracy.

Try Genre Genie for yourself, and see Tom’s GitHub for more details.