A predictive model for forecasting dental procedures, built during an ML engineering internship at Metasoft. The system processes over 6 million patient records to predict the most likely next procedures.
Approach
- Tested deep learning architectures: GRU, LSTM, and attention-based combinations
- Achieved 80% accuracy for predicting the top ten likely next procedures
- Trained on Google Colab TPU over 50 epochs
- Integrated with Metasoft's SQL Server infrastructure for production use
Outcome
The model enables clinics to anticipate resource needs and improve scheduling efficiency based on historical procedure sequences.