CV decision tree
Elderly people can fall down and there might be no help in remote access and/or no notification happening to send help.
What it does
This model can be a true super hero / life saver. It perfectly predicts whether a person is falling / walking using sensory data from a smartphone which could be used as a trigger to notify emergency services.
How I built it
Decision-tree based and cross-validated, feature-rich self-collected sensory data using Python for pre-processing and R for model training.
Challenges I ran into
- time inconsistent measurements of sensory data
- falling on the ground to collect data hurt sometimes
Accomplishments that I'm proud of
- identified most relevant features that distinguish between two classes
- showed that simple model can better predict and train much faster than a complex, deep learning model (LSTM)
What I learned
- feature extraction of time dimension
What's next for Human Activity Recognition
- collect more data to train more complex models
- incorporate more different activities
- integrate it to a server application that monitors live sensory input data and triggers a mechanism in case of dangerous situation