We got inspiration from 3Blue1Brown (Youtube Video Creator) who showed the beautiful complexity of machine learning in a very visually appealing and simplified concept video.
What it does
We can inputs data of University students in .csv data files that were expertly prepared for use by Accenture. We made a script to organize the data for the machine learning algorithms to be able to use them in calculations. After the calculations are done, the predicted drop-out status created by the bot and the actual drop-out status are compared. If it was determined that the machine was wrong, it will tweak the value of it's "weights" using simple algebra. The process will be looped over 30,000 times. By then, the algorithm will be able to correctly guess the drop-out rate up to 99% accuracy.
How we built it
We used Python to import the data from the .csv files to organize the data into a .pkl file. In order to save on limited memory, we saved the data into the .pkl file so that we can open and close the file on command without leaking too much memory.
Challenges we ran into
Although there were many, we'll only mention a few. Firstly, we were first going to use PNN (Probabilistic neural network) instead of the ANN(artificial neural network) that we used. The reason we made the switch halfway through was because the complexity of the PNN algorithm was too hard for us to properly code ourselves. Secondly, the data was way too big for us to effectively use to the full potential, so we had to select only a couple few which lead to many biases in the algorithms.
Accomplishments that we're proud of
It was our first time ever at a Hackathon. Also, it was our first time coding and researching a machine learning algorithm. As a team with only Freshman and Sophomores, we're very glad to see the project go through its final prototype stages even with such a difficult project for our skill level.
What we learned
We learned that there is not nearly enough time to finish writing a code to the fullest potential in just 36 hours. Mostly though, we probably researched machine learning on the web for around 8 hours in total before we started to even code. This was such a great experience to have an environment where learning in such intensities was not at all stressful but fun and enjoyable, especially because of all the other Hackathon members.
What's next for ANNIP (Artificial Neural Network In Python)
We will look into making sure that at least most of the negative biases associated with our processing of the data is removed or altered from the code. We also want to make use of the constantly changing variables (weights) in order to create a live graphing feed of these life-like growths of these neurons into their efficient values. Next, we want to try to use the same data on a PNN algorithm and see how it differs from an ANN especially on how well it performs in runtime and data processing.