Predictive Maintenance Data Scientist - Freelance
Anomaly detection for train maintenance systems. Developed ML models, conducted exploratory data analysis, and collaborated with domain experts to build internal tooling for reproducibility.
About this role
As part of the development of the internal application used by technicians to monitor train maintenance, RATP brought in data scientists to build anomaly detection algorithms capable of identifying the most relevant interventions among the large volume of fault messages sent by trains.
After an exploratory phase focused on analyzing the available data, I developed an anomaly detection algorithm designed to identify the most relevant fault codes. This required close collaboration with domain experts and end users to leverage their operational knowledge of train operations and maintenance.
Numerous workshops were organized with business experts to iterate on requirements and validate the approach. As part of improving and standardizing practices within the data science team, I also contributed to the development of an internal package enabling data scientists to analyze data and build models more quickly, robustly, and reproducibly — making it possible to deploy models more efficiently across new train lines and rolling stock.
Key contributions
- Developed an anomaly detection algorithm to identify the most relevant fault codes from train fault message streams.
- Conducted an in-depth exploratory data analysis phase before modeling, in collaboration with domain experts.
- Participated in workshops with business experts and end users to iterate on needs and validate technical choices.
- Contributed to an internal data science package to standardize and accelerate model development and deployment across new train lines.