This document discusses feature selection techniques for automatic emotion recognition (AER) systems. It describes how AER systems are divided into four sections: signal acquisition, feature extraction, feature selection, and classification. It evaluates different feature extraction and feature selection techniques for AER, including mel-scaled power spectrum, mel frequency cepstral coefficients, shifted delta coefficients, lasso regression, and ridge regression. It finds that combining features can increase accuracy but also dimensionality and computation time. Feature selection with lasso and ridge regression can overcome this issue while maintaining high accuracy levels. Testing on different databases yielded accuracy levels of 72.5% for SAVEE and 56.41% for eNTERFACE.