Bankrupcy Dependency visualized from Neural Net
We were interested in machine learning and data analytics and decided to pursue a real-world application that could prove to have practical use for society. Many themes of this project were inspired by hip-hop artist Cardi B.
What it does
Money Moves analyzes data about financial advisors and their attributes and uses machine's deep learning unsupervised algorithms to predict if certain financial advisors will most likely be beneficial or detrimental to an investor's financial standing.
How we built it
We partially created a custom deep-learning library where we built a Self Organizing Map. The Self Organizing Map is a neural network that takes data and creates a layer of abstraction; essentially reducing the dimensionality of the data. To make this happened we had to parse several datasets. We used beautiful soup library, pandas and numpy to parse the data needed. Once it was parsed, we were able to pre-process the data, to feed it to our neural network (Self Organizing Map). After we were able to successfully analyze the data with the deep learning algorithm, we uploaded the neural network and dataset to our Google server where we are hosting a Django website. The website will show investors the best possible advisor within their region.
Challenges we ran into
Due to the nature of this project, we struggled with moving large amounts of data through the internet, cloud computing, and designing a website to display analyzed data because of the difficult with WiFi connectivity that many hackers faced at this competition. We mostly overcame this through working late nights and lots of frustration. We also struggled to find an optimal data structure for storing both raw and output data. We ended up using .csv files organized in a logical manner so that data is easier accessible through a simple parser.
Accomplishments that we're proud of
Successfully parse the dataset needed to do preprocessing and analysis with deeplearing.
Being able to analyze our data with the Self Organizing Map neural network.
Side Note: Our team member Mikhail Sorokin placed 3rd in the Yhack Rap Battle
What we learned
We learnt how to implement a Self Organizing Map, build a good file system and code base with Django. This led us to learn about Google's cloud service where we host our Django based website. In order to be able to analyze the data, we had to parse several files and format the data that we had to send through the network.
What's next for Money Moves
We are looking to expand our Self Organizing Map to accept data from other financial dataset, other than stock advisors; this way we are able to have different models that will work together. One way we were thinking is to have unsupervised and supervised deep-learning systems where, we have the unsupervised find the patterns that would be challenging to find; and the supervised algorithm will direct the algorithm to a certain goal that could help investors choose the best decision possible for their financial options.