Linear Discriminant Analysis or LDA is one of many approaches used in supervised machine learning to solve multi-class classification problems. It uses many different methods from statistics and data processing to make conclusions about given criteria using the input data.
In this specific project, LDA is being used to classify various postures of hand movements based on Electromyography (EMG) data gathered from the forearms of multiple subjects. The data classified was gathered using 8 EMG electrodes measuring 11 postures over 6 trials. The raw data was taken and passed through our proprietary open source algorithm to classify each posture with an intermediate level of accuracy. Future improvements can be made to significantly increase the accuracy. Below is a detailed UML of how the program works.
This matrix shows the accuracy of our model by showing the number of correctly classified postures relative to the expected number for the repsective posture. For example, the TAB posture was correctly classified 80.25% of the time on the high end while HC was classified correctly only 38.07% of the time relative to the expected.