Pre-sleep stress has been shown to drastically lower the quality of rest, however, the specific changes in brain activity associated with it as seen through a single modality are not clear. Our study offers a thorough research and recognition system that uses multimodal EEG and EOG signals to measure how pre-sleep stress affects sleep architecture. In order to preprocess the Sleep-EDF dataset, we have applied a stringent processing pipeline that entails bandpass filtering and ICA-based artifact removal. We have devised 28 spectral and statistical features from the cleaned signals, such as relative band powers, Beta/Delta ratio, spectral centroid, and EOG-derived metrics, to represent cortical hyperarousal and sleep disruption. Our research reveals that pre-sleep stress results in a significant electrophysiological shift: an almost uniform 20–30% increase in Beta power and a quite significant increase in the Beta/Delta ratio, along with a 15–18% decrease in slow-wave Delta activity. These conclusions, as supported by spectrogram illustrations, tell the story of a stress-induced hyperarousal state that leads to the interference of slow-wave sleep necessary for recovery. Most importantly, we have put in place and assessed the performance of several machine learning models through this multimodal feature set for the binary classification of stress-affected sleep epochs. The ensemble models have reached a high level of classification performance; thus, they have been able to confirm the extracted biosignature’s discriminative power. The present study is an example of how the combination of multimodal EEG-EOG analysis and machine learning can serve as an efficient and objective means of quantifying and automatically detecting stress related sleep disturbances, thus it opens the door to embedded diagnostic tools in sleep monitoring systems.