We are trying to implement the P300 speller scenario. The Emotiv Epoc headset is being used to acquire the acquistion data. The data is then filtered and downsampled and fed to the classifier trainer which is trained accordingly, and a high percentage of classifier performance is achieved.
On moving on to the online scenario, the new is then fed into the voting classifier which uses our previously trained classifier processor, as per p300 online scenario. Unfortunately, the systems gives incorrect results, it either gives access letters as output or no letters at all.
The voting classifier has a parameter Number of Repetitions which on varying varies the number of output letters. But was it actually changing?
How can we improve our classifier performance to obtain good results?
Please help I am a newbie in this. Thnkx
P300 Online Speller
Re: P300 Online Speller
You should make sure that the number of repetitions in your P300 stimulator corresponds to the number of repetitions in your voting classifier if no averaging of the epochs is used. If you average all the repetitions, then you only need to vote once to get a result. If you have 12 repetitions and you average every 4 epochs, then you need to vote every 3 "repetitions". It seems you made some changes to some parameters in the scenario that are not consistent.