Ctrl Shift Elite

Inspiration

We are a group of NUS students who are highly interested in data analysis and how AI/ML can use the processed data to output accurate predictions.

What it does

Our project takes in company data and sales figures and carries out general preprocessing on this data. Our model is trained using this processed data, and can then predict sales figures given input of the similar format.

How we built it

Data manipulation and processing is mostly done by Pandas and Numpy in Python. Matplotlib was used for visualization of the given data. Scikit-Learn was used for k-means clustering to group company into regions. Model from XGBoost was used to train and predict the target values.

Challenges we ran into

Being unfamiliar with SIC Code, we had to read up more on the different categories of industries, and how we can use it to achieve more accurate sales prediction.

Accomplishments that we're proud of

We are proud of achieving a r^2 score of up to 0.98 on our test data.

What we learned

We learnt the importance of processing the data appropriately to achieve better predictions as inconsistencies in the data such as outliers can heavily affect the prediction outcome. We also explored different regression models such as Linear Regression, LASSO, Ridge Regression, Gradient Boosting Regression and ended up with XGBoost Regression which is a form of regularized Gradient Boosting Regression. From this, we learned that the basic models we learned in school such as the Linear Regression may not work so well on real life complex structured (or unstructured) data as data points may not be normally distributed.