WISJ ML Summer School

Project Overview

Duration: July 12th - August 23rd, 2025 Location: Code Chrysalis: Coding Bootcamp in Tokyo, Japan
Project Name: IntelliBS - Intelligent 5G Base Station Selection for Vehicular Networks
Technologies: Machine Learning, Random Forest, Feature Engineering, Python

Program Background

The WISJ (Women in Science Japan) Machine Learning (ML) Summer School is a selective six-week program designed to teach women and gender minority scientists applied machine learning through lectures, hands-on tutorials, and mentor-guided projects. The program aims to equip participants with the technical knowledge and skills needed to continue their machine learning journey in their respective scientific domains.

My Motivation

As a researcher who had primarily worked with machine learning in theory, I joined this program to gain practical, hands-on experience applying ML techniques to real-world problems. I was particularly interested in bridging the gap between theoretical knowledge and project implementation, especially as I was facing challenges in my own research that required a fresh, data-driven perspective.

Problem Statement

In modern vehicular networks, signal propagation poses significant challenges for maintaining reliable 5G connectivity. As vehicles move through urban environments, they encounter:

  • Dynamic blockages from buildings, obstacles, and other vehicles
  • Rapid mobility requiring frequent handoffs between the Base Stations (BSs)
  • Signal quality variations that impact communication reliability
Visualization of signal propagation in vehicular networks with Ray-Tracing

The critical question: How can we intelligently predict the optimal BS for a moving vehicle to ensure seamless connectivity?

Project Objective

The goal of IntelliBS was to develop an intelligent system that:

  1. Predicts the best 5G BS for vehicles in real-time
  2. Handles mobility and blockage challenges effectively
  3. Achieves high accuracy suitable for real-world deployment
  4. Provides immediate deployment value for next-generation connected vehicles

Methodology

Our approach combined feature engineering with machine learning:

  1. Feature Clustering: Analyzed and grouped relevant features affecting signal quality
  2. Random Forest Classification: Implemented an optimized Random Forest model for BS selection
  3. Performance Optimization: Iteratively refined the model through data analysis and parameter tuning
  4. Validation: Tested the system under various scenarios to ensure robustness

Results & Demonstration

The project successfully achieved near-perfect performance in predicting optimal BS selection. The final results were presented in a 3-minute presentation at the summer school.

Final presentation: IntelliRSU - Intelligent 5G Base Station Selection

Key Achievements:

  • ✅ Solved critical mobility and blockage challenges in vehicular networks
  • ✅ Achieved near-perfect accuracy through feature clustering and Random Forest optimization
  • ✅ Developed a framework ready for real-world deployment

Reflections & Key Learnings

This summer school experience was truly transformative for my research journey. Although I had theoretical knowledge of machine learning, this was my first opportunity to apply it to a complete, practical project from start to finish.

What made this experience special:

  • Problem-solving through different perspectives: When facing challenges in understanding the data, I learned to approach problems from multiple angles - observing patterns, questioning assumptions, and iteratively refining my approach. This process of discovery and problem-solving brought me genuine joy and excitement.

  • Invaluable mentorship: My mentor Alexey provided exceptional guidance throughout the project. His expertise was instrumental in shaping both the technical approach and the quality of this work. His advice helped me navigate challenges and think critically about model optimization.

  • Inspiring community: I had the privilege of meeting many talented women and professionals in the field. The collaborative and supportive environment fostered by WISJ created an ideal space for learning and growth.

  • Bridging theory and practice: Moving from theoretical understanding to hands-on implementation gave me confidence to tackle data-driven problems in my own research with a more practical, experimental mindset.

Future Directions

Future work will focus on:

  • Extending the framework to larger-scale networks
  • Investigating ensemble methods for even higher accuracy
  • Real-world field testing in diverse urban environments

Acknowledgments

I would like to express my sincere gratitude to:

  • Alexey, my mentor, for his invaluable advice and guidance throughout this project
  • WISJ AI Summer School for providing this incredible opportunity to develop cutting-edge research in an inspiring and collaborative environment
  • JST SPRING-GX program for supporting my participation