Feature Generation with Genetic Algorithms for Imagined Speech Electroencephalogram Signal Classification
Edgar Lara-Arellano, András Takács, Saúl Tovar‐Arriaga, Juvenal Rodríguez‐Reséndiz · Eng—Advances in Engineering · 2025
This work presents a method for classifying EEG (Electroencephalogram) signals generated when a person concentrates on specific words, defined as “Imagined Speech”. Imagined speech is essential to enhance problem-solving, memory, and language development. In addition, imagined speech is beneficial because of its applications in therapy fields like managing anxiety or improving communication skills. EEG measures the electrical activity of the brain. EEG signal classification is difficult as the machine learning (ML) algorithm has to learn how to categorize the signal linked to the imagined word. This work proposes a novel method to generate a specific feature vector to achieve classification with superior accuracy results to those found in the state of the art. The method leverages a genetic algorithm to create an optimal feature combination for the classification task and machine learning model. This algorithm can efficiently explore ample feature space and identify the most relevant features for the task. The proposed method achieved an accuracy of 96% using eight electrodes for EEG signal recordings.