1  Introduction

The transition to electric mobility is a cornerstone of global strategies to mitigate climate change and reduce dependence on fossil fuels. While the adoption of Electric Vehicles (EVs) has accelerated in recent years, the development of supporting infrastructure has often lagged behind. A primary barrier to mass adoption remains “range anxiety”—the fear of running out of power before reaching a charging point.

To address this, urban planners must move beyond reactive infrastructure development. The placement of charging stations is not merely a logistical challenge but a complex spatial optimization problem that requires balancing grid constraints, traffic patterns, and user convenience. As noted by Knuth in his seminal work on algorithms, optimization often requires a systematic approach to handle complexity Knuth (1984).

This project proposes a data-driven framework to solve the “optimal location” problem, leveraging machine learning to predict high-demand zones with precision.

1.1 Overview

This project integrates diverse geospatial datasets to identify the most strategic locations for new EV charging stations. Unlike traditional methods that may rely on intuition or simple heuristics, our approach utilizes a Machine Learning (ML) pipeline to quantify demand at a granular level.

We employ an XGBoost Regressor to model the relationship between urban features (such as traffic volume, population density, and Points of Interest) and charging demand. The output of this model is then refined using a Multi-Criteria Decision Making (MCDM) technique, which ranks potential sites based on both predicted demand and land suitability. The final output is a prioritized list of locations that maximizes utility for EV owners while ensuring economic viability for operators.

1.2 Existing System

Current methodologies for planning EV infrastructure often suffer from significant inefficiencies. The existing “system” for site selection typically falls into one of three categories:

  1. Opportunity-Based Planning: Stations are installed wherever land is cheapest or where government grants are available, rather than where drivers actually need them.
  2. Reactive Deployment: Infrastructure is added only after significant congestion or user complaints arise in a specific area, leading to a perpetual lag behind demand.
  3. Manual Heuristics: Planners use basic spreadsheets or simple proximity rules (e.g., “place one station every 5km”), ignoring complex non-linear factors like traffic flow direction, “point of interest” clustering, or grid load capacity.

These approaches result in a polarized network: some stations are perpetually overcrowded, while others sit idle in “charging deserts,” wasting capital resources.

1.3 Objectives of Project

The primary goal of this project is to replace subjective planning with a robust, quantitative framework. The specific objectives are as follows:

  1. Data Integration: To aggregate and preprocess heterogeneous geospatial data—including road density, traffic volume, demographics, and POI distribution—into a unified analytical dataset.
  2. Demand Modeling: To develop and train a Predictive Machine Learning Model (XGBoost) capable of estimating the potential charging demand score for any given grid cell with high accuracy.
  3. Feature Importance Analysis: To interpret the model results and identify which urban factors (e.g., commercial density vs. residential density) are the strongest drivers of EV charging demand.
  4. Strategic Optimization: To implement a Multi-Criteria Decision Making (MCDM) system that filters and ranks the top candidate locations, delivering a final list of the Top 300 Priority Sites for immediate infrastructure development.