Clean & tokenize
Applied text mining and natural language processing to clean, normalize, stem, and tokenize the input data.
Case study · AI & machine learning
Machine Learning
for Safety Reports
Turning safety reports into structured data.
KBD explored speech recognition, natural language processing, and machine learning to help categorize thousands of handwritten safety reports.
The workflow
A path from reports to usable data.
- 01 / InputSafety report data
- 02 / ProcessText preparation & classification
- 03 / OutputCategorized reports
- Focus
- Safety report classification
- KBD contribution
- Data preparation & model training
- Delivery
- Proof of concept
The challenge
Less transcription.
More useful data.
Safety reports were recorded by hand, transcribed into computers, and then analyzed manually. Processing thousands of reports this way took time and left room for inconsistent results.
The task was to explore how speech recognition and machine learning could turn those reports into accurately categorized data.
KBD’s contribution
From raw text to a trained model.
Classify & predict
Explored semantic analysis, text and word classification, and supervised learning to compare training data with predictive results.
Train & evaluate
Trained and evaluated the model over successive iterations, using the results to improve its classification performance.
Project outcome>0.95Reported F1 score
The outcome
A foundation for further development.
KBD’s original project account reports an F1 score above 0.95 after iterative model improvements. F1 combines precision and recall to measure classification performance.
The result established a proof of concept for larger implementations, showing how text classification could support a workflow that previously relied on manual processing.
Your next project
Make more of
your data.
Tell us about the documents, workflows, or decisions you want to improve.
Start a conversation