My 1000 dollars Story
It was the 11th of October, which happened to be my birthday. A task was given and the task was to predict if a certain loan (from One-fi to a customer) would be good or bad based on Customer credentials.
In order not to bore you I’ve decided to run through the important stuff.
This was how it went down.
At first, the task looked familiar and easy but I just couldn’t figure it out for quite a while, and then boom, I figured out the whole task…
We were given 3 different datasets for both Train and Test. One containing the demographics of each customer, another containing information on customers’ most recent loan, and lastly the performance (good or bad) of the last dataset containing information about all previous loans customers had prior to the one on the performance data.
The first step I took was to find out what it takes for a loan to be a good one or a bad one.
So, after all the research and reading the task started;
I started by reading all the 7 datasets with pandas, then I merged each train and test data
I extracted lots of features from the previous loan, and demographics data. It wasn't an easy task, stayed up all night learning and implementing. After all the feature extraction , I merged the extracted features from previous loan and demographics data to the performance data which the results is shown below;
Doing all this took most of my time, especially my nights.
The most important part of the task was picking the right algorithm.
“To be honest i don’t really know much about all this algorithms, i just try stuffs until it works”
So, I tried GradientBoostingClassifier first, the score was really bad it was below SampleSubmission score.
The result wasn't an encouraging one, I had to go rest my head.
After sleeping for 2 hours, I tried the second algorithm — Xgboost Classifier (Extreme Gradient Boosting Classifier) since GradientBoosting didn't go too well, then something extreme was tried. The result was extreme but also wasn't that good.
Guess I had to make some changes to the Xgboost parameters but things only got worse.
I was a bit frustrated. Two days gone! The first algorithm gave me a score less than the sample submission, however, the second was slightly better. So, I asked myself, is this the best I can do? I was about giving up but I decided to give it my last shot.
Azure ML came to mind. Azure ML is a tool for performing machine learning tasks online. It has many algorithms and was developed by Microsoft. On Azure ML, I used Two Class Neural Network, and the score was quite impressive to me
The score from the Neural Network was the winning score, it came at the right time when I was about giving up.
I think the lesson here is that one shouldn’t give up easily. After every disappointment, pick yourself up and keep going, try again making sure your next trial is better than the previous.
And that is how I won the 2018 Data Science Nigeria (DSN) hackathon.
My appreciation goes to Data Science Nigeria, Bluechips Technologies, Tejumade Afonja and the AI Saturdays Lagos folks, my friends and family.
My name is Sadiq Aderinto and this is my thousand dollar story.
