Smart Smoke Alarm ML Analysis
(2023)Fire Safety Enhancement Research
Duration: 6 weeks
Synopsis
Machine learning research to modernize smoke alarm technology. Evaluated 5 algorithms on 60,000+ IoT observations to reduce false alarms while eliminating failures in detecting real fires.
The Making Of
🎯 The Challenge
Traditional smoke alarm technology unchanged since 1970s. In 2015, Australian false fire alarms were 97%, while ionisation alarms had 55% failure rate in real fires. This creates public complacency.
💡 The Solution
Analyzed 60,000+ IoT observations including temperature, humidity, CO₂, and particle size. Tested 5 algorithms: KNN (weighted/unweighted), Naive Bayes (Multinomial/Gaussian), Logistic Regression, and MLP. Used Random Forest for feature selection and 70-30 cross-validation.
🚀 The Impact
Weighted KNN achieved 99.97% accuracy with only 0.04% false negative rate (missed 2 fires out of 15,000 tests). Neural network had dangerous 13.95% false negative rate. Findings suitable for industry pitch to Bosch, Honeywell, ABB.
Key Metrics
Tech Stack (Cast & Crew)
- PythonImplementation
- Scikit-learnML Framework
- Random ForestFeature Selection
- KNNBest Performer
- Google ColabCompute
Quick Info
- Year
- 2023
- Duration
- 6 weeks
- Complexity
- Mid
- Featured
- No