Navigating the confusing world of credit scores and loan payments can be overwhelming for any of us. But add the extra financial burdens often shouldered by the “New Middle Class,” and a huge opportunity to use technology for good emerges.
score is a native iOS application that provides individuals with a holistic “risk score” from aggregated demographic, financial, health, social, and professional data. The goal is for score to inform enable a better lifestyle -- from exercise and nutrition to social media behavior to spending habits. While financial institutions have traditionally relied on demographic and financial data to determine the risk level of loaning to a certain individual, our application and AI model take advantage of less conventional sources and types of data -- health, social media, and professional activity.
What it does
At its core, score uses data from four major categories in addition to basic demographic data: health data (Nokia Health), social media data (Facebook), professional data (LinkedIn), and financial data (financial institution). The health data is further supplemented by a nutrition documentation feature using Apple’s CoreML Vision classifier. We used machine learning (Python scikit-learn) to create the model and trained it on our dataset using a Random Forest Classifier.
Upon logging in, the app tells the user what his/her risk score is -- the higher the score, the better (aka the less risky he/she is to financial institutions). The user can then choose to explore the components of his/her holistic score, view his/her profile, and potentially chat with a chatbot to receive personalized recommendations and suggestions. To supplement the health data, score uses computer vision to classify the nutrition score of the user's meals. Ultimately, our application is meant to smoothly guide the user towards a happier, lower-risk lifestyle.
How we built it
We used Xcode and Swift to build the iOS application. The app also uses Apple’s CoreML framework to tap into Vision and image classification. Our app is connected to a Flask backend running on Heroku and Docker.
Our "riskiness" dataset was gathered from Kaggle datasets and supplemented with our own research on loans and loan repayment behavior. Some of the data included is as follows: riskiness, education level, gender, money in checking account, money in savings account, credit history, credit score, years at present job, years at current address, currently employed, number of Facebook friends, Facebook sentiment analysis, average daily steps, average daily intensive activity, nutrition score, number of LinkedIn connections.
Challenges we ran into
Learning new things and staying up late is tough as always. It was also somewhat difficult to track down data regarding loans and build a comprehensive dataset.
Accomplishments that we're proud of
We successfully integrated APIs into our iOS app -- something we haven’t completed before! We also attempted machine learning and created a model with at least 70% accuracy.
What we learned
Machine learning, AI, iOS SDKs, all kinds of neat tech.
What's next for score
We’d like to add more functionality to score, including the potential to partner with a financial institution to provide users with real-time advice and communication (e.g., via phone or messaging) from representatives at the financial firm or through a chatbot. We would also like to continue building out the dataset and adding more streams of data to create a truly holistic view of an individual’s risk-inducing and risk-mitigating behaviors (e.g., sleep patterns, driving habits, risk scores of friends and family, etc.). There’s also an opportunity for score to scale up by allowing users to opt-in to providing data to financial firms as validation of their reliability. Finally, a social good application of score would be aiding at-risk individuals in the community to understand their own behaviors and navigate the steps to building a less risky future.