We were motivated by a general lack of personalization when it comes to investment advice for retail investors, as more often than not, the financial advisors would recommend the top-performing investment instruments to the client, who may or may not be familiar with that particular industry. We felt that this lack of personalization takes away the sense of control that the client has over his or her investment.
What it does
WalletStreet is a web-app that analyzes the user's purchase history and spending patterns, and identify an industry that the user is most likely going to have expertise in. The app then recommends to the user top-performing stocks in that particular industry, ranked from most desirable to least desirable based on key metrics and performance data.
How we built it
Our app is hosted on Heroku and Firebase. The back-end is programmed in Python and can be roughly broken down into three parts: user-account generation using Capital One's API, algorithms that process the user-account info and detect spending patterns and market segmentation, and stock performance analysis using our own weighted ranking systems and BlackRock's API.
We spent a good amount of hacking hours designing our supervised learning algorithms. After testing out multiple models including Naive Bayes Classification and ML relevance-scoring, we decided that it's best to use data normalization and vectorization as the main parts in our algorithm (along with other algos), since the accuracy of an unsupervised learning algorithm depends heavily on the number of iterations and training data sets. We generated over 17,000 entries as our training data set, which was ideal enough given our limited time and data availability but not ideal enough for unsupervised learning. Thus we used more supervised learning to guarantee precision and to have a better idea of how our algos work.
Challenges we ran into
Wiring our app together, generating user data with Capital One's API, algorithm designing.
Accomplishments that we're proud of
We are very proud of the complexity of our data analysis and how much we learned over the past 36 hours. Our team was beyond excited to find our top recommended stocks closely match the market recommendations. Furthermore, we blind-tested a biased data set and the result industry actually matches the bias in the data. Financial advising is a billion-dollar industry with numerous experts doing it as their day-job and so is machine learning, both of which we accomplished (to a decent level) over a weekend.
What we learned
We learned how to strategically analyze our data so that it works magic.
What's next for WalletStreet
The potential for this app is definitely beyond what we have accomplished over the past 36 hours. We designed the app with the key concepts of maintainability and adaptability in mind, and we believe that given more time and data sets, we could make WalletStreet even more accurate and personalized.