Synopsis
The final submission file can be found here
Inspiration
How much gas will be needed in the coming year? It's critical for Chevron to have an understanding of how much gas they will need to provide in order to ensure they have enough (but not too much) to satisfy the demand. In order to do so, we have to understand how many vehicles and the types of vehicles that we expect will be on the road. This project aims to predict vehicle inventory based on historical data,
What it does
Our model aims to take in a variety of features on a vehicle and predict its vehicle inventory in the year of 2024.
How we built it
My teammate and I used tree-based models to analyze and predict the data. Since the dataset contains mostly categorical variables, we first converted them to dummy variables. We also added the difference between date and model year as an additional continuous variable. We first trained a Random Forest model to our dataset, and the performance was moderately good. We then trained an XGBoost model which achieved a better performance. In the end, we decided to go with the XGBoost model as it achieved a lower RMSE value.
Challenges we ran into
One challenge we ran into was that we spent a large amount of time on data cleaning, and we also tried to do data imputation on some missing values. However, the model performance wasn't as good as we expected. This could be due to the randomness of the noise or the outliers that potentially arose from our imputation.
Accomplishments that we're proud of
We are proud of finishing the project and achieve a relatively low RMSE, by using the models that we weren't 100% familiar with previously.
What we learned
We learned that data cleaning and preparation is a critical and long step in the process, and that the actual modeling is rather quick once the data is fully cleaned and ready.
Log in or sign up for Devpost to join the conversation.