EEG(electroencephalogram) measures (volts) electrical activity generated by the activity of neurons in the brain. Waves at specific frequency patterns are examined for arriving at results.We are trying to solve the existing problems in the classification of brain signals to digits (09). We aim to classify digits on a standard dataset named “MindBigData”.Our proposed approach is to fit the available models with machine learning methods like LSTM(LongShort-Term-Memory) and SVM(Support Vector Machine) to attempt in increasing the accuracy obtained in data analysis and digit prediction
Using LSTM, instead of transforming the data completely, it only makes small modifications, selectively remembering or forgetting things. Using SVM, it can try to achieve a good margin for the separation of the different classes (digits) for improving accuracy. This can be done by tuning certain parameters.
Upon fitting the model with neural networks using Long Short Term memory and Support vector machines, we have classified the EEG signals into digits on a standard dataset named “MindBigData”. This involved tuning of parameters and various changes in the neural network model. Frequent comparison of scores were taken and the best accuracy upon a particular change was noted. We have used Keras neural network python library for implementing the code
We aim to generate our own dataset for this study using a headset and EEG signals . We also aim for trying different types of models and work on further improving the accuracy in classifying the digits more effectively. Different scores can be shown to compare recent results with previous ones, arriving at meaningful results. We also want to further use EEG signals for other classification problems such as alphabets and direction.
Download the dataset from "https://github.com/meagmohit/EEG-Datasets".