Driver fatigue is a major cause of traffic accidents. Electroencephalogram (EEG) is considered one of the most reliable predictors of this cerebral state. This paper proposes a novel method for detecting fatigue from a single electrode, that can be implemented in real-time. The paper first presents a method to determine the most relevant EEG channel to monitor fatigue, using maximum covariance analysis. The second contribution is developing a machine learning method to detect fatigue from this single channel using spectral features and a Long Short-Term Memory (LSTM) deep learning model. Experiments with 12 EEG signals were conducted to discriminate the fatigue stage from the alert stage. Our main finding shows that TP7 was the most relevant channel for monitoring fatigue. Interestingly, this channel is in the left tempo-parietal region where spatial awareness and visual-spatial navigation are shared. It is also related to the faculty of cautiousness. In addition, despite the small dataset, the proposed method yields 75% accuracy for predicting fatigue with a 1.4-second average delay. These promising results provide new insights when developing strategies for monitoring driver fatigue.