Swiss public transport network delay predition (imbalanced dataset)

Capstone project for “Applied Data Science: Machine learning” course from EPFL in Lausanne

Project presented by EPFL during “Applied Machine Learning Days 2021: Impactful Projects Brought to Life” session - starting at 30:30

Context

Switzerland public transports are known worldwide for their punctuality. I wanted to confirm this hypothesis that overall public transport are reliable while simulatenously assesing potential pockets of lower performance (e.g.: providers, stations or mean of transports with frequent and/or important delays)

Objectives

Primary: predict delays for train connections in swiss transport network
Secondary: assess transport network performance and identify regions/providers/lines with more frequent delays (part of EDA)

Content

• Project proposal including EDA: “Capstone-proposal-GregoireLaurent-vFinal”
• Data cleaning and engineering based on EDA: “Capstone-project-GregoireLaurent-Datacleaning”
• Comparion of ML models and final conclusion: “Capstone-project-GregoireLaurent-MLtasks”
  • ML models including: Logistic Regression, SVMs, Trees and Keras
  • Performance assement based on F1 scores and Area under Precision-Recall curve

For more details see GitHub Capstone project - Gregoire LAURENT.