This week I have been continuing work on the machine learning aspect of the signal processing. I obtained my first accuracy result from splitting our 42 trials of data between training, validation, and testing data with an accuracy of 35% using a Long Short-Term Memory (LSTM) algorithm classifier. This definitely does not meet the accuracy we are aiming for so there is still work to be done in the adjustment of hyperparameters and the input feature vector.

In my further exploration of the data, I resolved to filter the data with a bandpass filter between 50 and 150 Hz in order to remove noise and artifacts from the data. This significantly smoothes out the curves as we can see in the figure below.

I am continuing to work with PCA in order to reduce the dimensionality of the data and get a meaningful eigen decomposition of the signals in order to include in the feature vector.

 

The plan for the upcoming few days before Thanksgiving is to improve the classification accuracy and develop a robust feature vector in order to come together with the rest of my teammates through our final integration phase. This would leave us on track in order to complete all the deliverables for the final project.