BeliEVe

BeliEVe

"Believe you can and you're halfway there." - Theodore Roosevelt

Have you ever wondered what the population of vehicles might look like in the future? Well, look no further. BeliEVe does the magic for you!

What does it do?

BeliEVe relies on the good old random forest model to train over data from 2019 to 2024 (based in California) to predict the vehicle population in 2025.

How we built it

We built our model in three stages:

Pre-Processing

• Check for NaN values within the dataset, and we used MICE imputation to fill in the missing data
• Changed data types of certain columns that had mismatched

Feature Engineering

• Used Correlation Matrix and Random Forest Feature Importance to single out key features that strengthen the correlation

Visualize

• Visualised key metrics and impactful insights from the dataset to reinforce market trends for cars for both gas or Electric Operated Vehicles.

Training the model

• Scaled the model using log transform during training and brought back the original value by transforming it back to exponential and compared RMSE with testing results.

Here are the benchmark

Challenges we ran into

Initial Large RMSE due to the target variable's skewness really tired us.

Accomplishments that we're proud of

Getting a good R2 Score, probably.

What we learned

How to categorize and treat

What's next for BeliEVe

Make it a production ready model for Chevron

Built With

• numpy
• pandas
• python
• scikit-learn
• seaborn
• statsmodels