FinSightAI

• 

  Preprocessing Page

Inspiration

When the finance theme was announced we realized both had experiences with banks taking a long time to process different kinds of applications. From there we did research into risk analysis and how small banks don't have the capability to create or maintain such models. The lightbulb moment was when we thought about tools like PowerBI that seamlessly helped people conduct data analysis without having to know about the inner workings of it and FinSightAI was born.

What it does

Our application automatically completes preprocessing steps for hundreds of thousands of rows in a matter of seconds, building reliable and accurate AI models which can then be scaled and exported based on business needs. Our powerful preprocessing tool makes it easy and intuitive to wrangle data, allowing clients to easily manipulate/delete columns, implement feature selection, automatically encode categorical data, normalize/scale prior to training, and even detect unbalanced datasets, employing synthetic minority oversampling techniques to make training data evenly distributed. FinSightAI uses built-in intelligence to detect if your model is overfitting data, wasting time on unnecessary training, and when it's ready for real-world predictions. This helps non-technical stakeholders understand limitations and helps technical stakeholders save time on repeated processes.

How we built it

Google Colab was used as an IDE to build and test our built-in intelligence model which was made with Python. Python was also used for the UI and the backend. Additionally, we used libraries such as customtkinter, numpy, matplotlib, pandas, seaborn, CTkMessagebox, scikit-learn, and imblearn.

Challenges we ran into

The visualization for the training/validation was not updating in real time after every epoch that the model trained. We spent over 1.5 hours trying to fix this issue, eventually realizing that it was a small bug. Another challenge that we ran into was deciding the scope of this project. We could have gone down the route of turning this into a learning tool for others or specializing it for non-technical audiences. After much deliberation and a mentoring session, we decided on this implementation.

Accomplishments that we're proud of

We are proud of being a team of two people who were able to go from ideation to finished product in under 12 hours.

What we learned

In terms of technical skills, we were able to learn a lot more about the inner workings of ML models and identifying and fixing issues in real time. In terms of the soft skills, we learned about effective communication and collaboration since we were working on different components of the project but making sure that when combined, they fit together seamlessly.

What's next for FinSightAI

We would like to give the user more customizations including different scaling options and models. We would also like to expand the accessibility of different tools like loan approval or portfolio management.

Built With

• colab
• ctkmessagebox
• customtkinter
• imblearn
• matplotlib
• numpy
• pandas
• python
• scikit-learn
• seaborn
• sklearn