Predicting Periodontal Disease with ML

Overview

This project studies whether breath-based VOC sensing and machine learning can support earlier and more accessible periodontal screening.

Key Focus

• Build a non-invasive screening path for periodontal severity outside a traditional clinic visit.
• Combine sensor signals with machine learning to identify risk patterns in a practical workflow.
• Test whether the approach still performs when moved beyond retrospective records.

Methodology

• VOC breath sensor measurements were paired with clinical and engineered feature sets for model development.
• Multiple machine learning models, including RF and SVM, were evaluated inside a structured validation pipeline.
• An external breath-testing protocol was used to check how well the approach translates to new participants.

Key Outcomes

• The external validation cohort reached 0.91 accuracy with an AUC of 1.00 in a small sample.
• Breath-derived signals showed enough structure to support meaningful severity classification.
• The results support the feasibility of lower-cost, non-invasive periodontal screening workflows.