KPMG-Sustainable-Future

MEMBERS: Alice Liu, Shoya Dixon, Wang Xiang, Melina Tsai, Rachel Tong, Ana Maria Rodriguez

---

CHALLENGE:

Identify and to provide a prescribe placement of electric vehicle charging stations (EVCS) that appealss to the private corporation persona within the Dallas, Texas area.

---

CONTEXT:

President Biden signed the bipartisan infrastructure bill in November 2021, including a $5 billion investment in self-administered grants for nationwide electric vehicle (EV) charging stations. Specifically for Texas, they plan to dedicate $400 million to implementing charging stations. Moreover, KPMG clients seek advise on how to identify where EVCS should be placed for maximum return on investment.

---

APPROACH:

1. Research

   • Brainstorm and research features that would help indicate where EVCS should be placed
   • Utilized Google, Kaggle, US Government website, KPMG provided data websites, OpenStreetMaps, Statistica, City of Dallas GIS Services, etc.

2. Data Visualization and Cleaning

   • Visualize what information the data is trying to tell us
   • Drop/replace missing values
   • Drop irrelevant columns
   • Rename columns
   • Convert rows to zip-code level

3. Feature Selections

   • All the data are in numerical value.

Demographics	Geographic Features	Economic Background
Population	Business/Residential building ratio	House Value
Gender	Shops	Household Income
Race & Ethnicity	Parking Lots	EV registrations
	Gas Stations

4. Data combination

   • Combined all the data through zip-code level

5. Modeling

   • Used: OLS, Lasso, Random Forest, Gradient Boosted Decision Tree (GBDT) model.
   • Evaluated each model using: RMSE, MSE, K-Fold Validation
   • RMSE yielded the best result
   • The lower the RMSE value it is the better the evaluation: RANDOM FOREST is the winner!
   • RMSE result:

Model Name	Description	RMSE
OLS	Fits a hyperplane by assigning optimal weights to each feature	191
Lasso	OLS that penalizes features with large weights	117
Random Forest	Fit multiple “overfit” trees in parallel and average the results	75
GBDT	Iterative tuning of a tree by correcting errors	79

6. Forecasting Zip Code Charactistics Used historical GDP data for Dallas 2002-2020 and ARIMA to help predict the FUTURE GDP data for Dallas 2020-2030. Utilizing these predictions, we obtain the top zip codes that needs EVCS. We sort by largest percentage.

   • The top 3 zip codes to place EVCS: 75246, 75210, 75253

---

SUMMARY

The supply and demand of EVCS can be viewed by examining the current EVCS available (supply) and EV registrations (demand)

The most important features (in descending order):

Features
Number of businesses
Asian population
Household income
Parking Lots
House value
Hispanic Population
American indian population
Other population
Black population
Population
Residential business ratio
Hawaiian population
Female population
Gas stations
Shops

The top 3 zip codes to place EVCS: 75246, 75210, 75253

For more in depth insight, refer to the following slide link for the Appendix: https://docs.google.com/presentation/d/1G_5Mfh4qrS0M1IgiVSp9ZvnUrjJShmQMcsFv2btxDrs/edit?usp=sharing

---

FUTURE GOALS

1. Prediction Improvements
   • Increase accuracy within the simulation model by adding randomization of percentage growth to each feature and using Census data for demographic features
2. Depth
   • Identify specific locations rather than higher-level zip codes
3. Scope
   • Expand analysis outside of Dallas:
     • Confirm analyses about predictors for EV Demand
     • Compare EVCS Supply and Demand in different regions