Recent advancements in plant electrophysiology suggest that plants can serve as biological sensors and are capable of detecting environmental and physiological signals. In this study, we explore the novel idea of using basil plants to classify human sleep stages based on their electromagnetic responses. We collected plant signals using low-cost Oxocard microcontrollers and paired them with sleep stage labels derived from the Sleep Cycle app. Two machine learning approaches were applied: a multi-class classification model that segmented the entire night into hourly intervals labeled as wakefulness, light sleep, or deep sleep, and a binary classification model that focused on the transition into sleep by labeling data before and shortly after sleep onset as awake or asleep, respectively, using 30-s segments. Mel spectrograms and MFCC features were extracted from the plant signal recordings to train deep learning models such as EfficientNetV2 and TinyViT. While the multi-class model achieved moderate accuracy (45.19%), the binary classification model performed significantly better, reaching an accuracy of 69.22%. Despite certain limitations including small sample size and potential inaccuracies in sleep stage labeling, our findings indicate that plant-based sensors may capture meaningful patterns related to human sleep, offering a low-cost and non-intrusive alternative for sleep monitoring.