machine learning - Classifying human activities from accelerometer-data with neural network -
I have been tasked to complete the existing classifier benchmark for my company. There is a difference between transport, like I'm currently on a train, driving or cycling, so this is the main focus.
I am reading a lot about LSTM, and its recent success in handwriting and speech-recognition, where the time between important events can be very long.
So, my first thought about the problem with train / bus is that it is probably not that for example, a clear and concise cycle as a walking / walking event is probably important.
Has anyone tried anything like decent results? Or are there other techniques that could potentially better solve this problem?
I have worked on mode of transport Using Smartphone Accelerometer I have found that the main result is That will make almost any classifier; The main problem is the set of features (this is not different from many other machine learning problems.) My feature sets with both time-domain and frequency-domain values taken from time-series sliding-window segmentation have ended.
Another problem is that the accelerometer can be placed anywhere on the body, it can be anywhere and in any orientation. If the user is running, then in a bag, in a car seat, a suction-cup window mount etc. is in the phone pocket?
If you want to avoid these problems, use GPS instead of accelerometer. You can do relatively accurate classification with that sensor, but the cost is battery usage.
Comments
Post a Comment