Predictive Analysis of EV/PHEV Sales for Dealerships

Project Overview

This project aims to analyze historical sales data of Electric Vehicles (EVs) and Plug-in Hybrid Electric Vehicles (PHEVs) to predict future sales trends. The analysis provides insights into growth patterns, market dynamics, and infrastructure needs, aiding a dealership in strategic planning and investment decisions.

Data Sources

The study utilizes data from the following sources:

• US Department of Energy
• Local DMVs of California, Florida, New York, Texas, and Washington
• U.S. Census Bureau
• U.S. Energy Information Administration
• Bureau of Labor Statistics
• The World Data Bank

Data Cleaning and Wrangling

Cleaning and wrangling the data involved:

• Focusing on a 10-year period for state-specific data
• Using recent data for U.S. general data
• Ensuring data was cleaned and processed for predictive analysis

Notebooks

1. Machine Learning Predictive Analysis.ipynb
2. States Notebook.ipynb

Machine Learning Predictive Analysis

Contents

1. Data Loading and Preparation

   • Loading the dataset
   • Converting 'Date' column to datetime format
   • Setting the date column as the index
   • Cleaning 'Gas Vehicles sales' column

2. Data Visualization

   • Visualizing trends for Hybrid/EV sales, gas vehicle sales, gas prices, diesel prices, unemployment rate, and inflation rate

3. Correlation Analysis

   • Calculating and displaying the correlation matrix

4. Feature Engineering and Normalization

   • Adding polynomial features
   • Normalizing the data

5. Model Training and Tuning

   • Defining models (Random Forest, Gradient Boosting)
   • Hyperparameter tuning using GridSearchCV

6. Ensemble Modeling

   • Creating and evaluating an ensemble model combining the best-tuned models

7. Feature Importance and Predictions

   • Visualizing model predictions vs. actual values
   • Plotting feature importances for both Random Forest and Gradient Boosting models

States Notebook

Contents

1. Data Loading

   • Importing state-specific data from various sources

2. Data Cleaning and Processing

   • Cleaning and wrangling state-specific data for analysis

3. Exploratory Data Analysis (EDA)

   • Visualizing state-specific trends and patterns

4. Model Development

   • Developing and tuning models for state-specific predictive analysis

5. Model Evaluation

   • Evaluating model performance using appropriate metrics

Conclusion

The predictive analysis indicates a strong growth trajectory for EV/PHEV sales, with estimates between 200,000 and 500,000 units. For the dealership, this suggests a significant opportunity to expand EV/PHEV inventory and enhance charging infrastructure. By leveraging these predictions, the dealership can capitalize on market trends, ensuring long-term growth and leadership in the EV market.

Dependencies

• pandas
• matplotlib
• seaborn
• scikit-learn
• xgboost
• catboost
• statsmodels

Install the dependencies using pip:

pip install pandas matplotlib seaborn scikit-learn xgboost catboost statsmodels

Author

Gian F. Villafañe Morales - Data Analyst