Lighting-pole Maintenance Prioritization

What it is

The project used condition data to help a startup prioritize lighting-pole maintenance. I cleaned and prepared the datasets with Python and RapidMiner, trained and compared several models, and produced two final models.

The problem

A startup needed to decide which lighting poles to maintain first based on available condition data, with limited local training data.

My role

Software Engineer

What I worked on

Cleaned and prepared condition datasets with Python and RapidMiner.
Trained and compared several machine-learning models.
Produced a model reaching 76% accuracy on a small local dataset.
Produced a separate model reaching 96% accuracy on a larger dataset with more features.
Communicated model results and limitations to a non-ML stakeholder.

Decisions and trade-offs

Separate dataset evaluations

The 76% and 96% results came from different datasets with different sizes and feature sets. They are not directly comparable as a before/after improvement.

Testing and verification

Compared multiple models on each dataset before selecting final candidates.
Documented dataset differences alongside accuracy results.

How it fits together

Condition data

Cleaning & features

Model compare

Maintenance priority

76% accuracy

Small local dataset

96% accuracy

Larger dataset, more features

Different datasets — not directly comparable