Stay well with machine learning
1 like1,013 views
This document discusses how machine learning can help people stay healthy by analyzing biometric data from wearable devices. It explains that measuring factors like temperature, heart rate, and activity level can provide insights to form good health habits. Machine learning algorithms like K-means clustering are applied to this biometric data to recognize patterns and activities. While a basic sample model is created, the document outlines ways to improve it, such as using more data, other algorithms, and ensembles of models. Overall, the document demonstrates the potential of machine learning and quantified self-tracking to provide health and wellness insights.
Stay well with machine learning
- 1. Stay well with machine learning Anastasiia Kornilova,! SoftServe Data Science Group! "1
- 2. Life expectancy at birth "2
- 3. Medicine evolution Fighting against diseases! Monitoring for 鍖nding early symptoms! Preventing diseases /Forming good health habits "3
- 4. Healthy habits Sleep! Running! Swimming! Walking! Diet! "4
- 5. How measurements can help form good health habits? Observer effect: Measuring system changes the system! Gami鍖cation! Psychological effect! Power, will, willpower "5
- 6. What we can measure? Easy/Relatively Easy:! Temperature! Weight! Heart rate! Blood pressure! Steps ! How about:! Activity level! Sleep quality! Predicting some diseases "6
- 7. Gadgets examples Fitness trackers/ Smart watches:! for every day usage! for measuring sleep quality! for trainings! Smart scale! Smart toothbrush "7
- 8. Life is short. Do things that matter http://mytikker.com/ "8
- 9. What is machine learning? "9
- 10. ML algorithms Supervised! Unsupervised ! Semi-supervised "10
- 11. How machine learning can help? "11
- 12. Data: Basis Band metrics Optical blood 鍖ow sensor - for measuring heart rate! Accelerometer - for measuring steps amount! Body temperature sensor! Galvanic Skins response sensor "12
- 13. Data "13
- 14. "14
- 17. "17
- 18. How good is our model? "18
- 19. "19
- 20. Ways to improve Improve current model: ! Data scaling! More data! Imputing! Use different model:! Supervised! Unsupervised! Use ensemble of models "20
- 21. "21
- 22. You can use ML algorithms for Fraud detection! Computer vision! Predicting future! Recommend product/friends! Speech and handwriting recognition! Identify key topics/summarize text! Find patterns in users behavior/actions "22
- 23. Summary We built simple 束toy損-model for activity recognition! Output of our model is relevant to to output of existing full-featured commercial service
- 24. Q & A "24