Machine Learning Distinguishes Plant Bioelectric Recordings with and Without Nearby Human Movement.

Area of Science:

• Plant electrophysiology
• Bioelectrical signaling in plants
• Human-plant interaction

Background:

• Detecting plant bioelectric differences near human movement is difficult due to low-amplitude signals and environmental interference.
• Previous research has not quantitatively assessed plant responses to proximate human activity.

Purpose of the Study:

• To quantitatively assess if plants exhibit measurable bioelectric differences in response to nearby human movement.
• To explore the potential of machine learning in distinguishing plant bioelectric signals under different proximity conditions.

Main Methods:

• Recorded bioelectric activity from 2978 plants (basil, salad, tomato) using leaf and soil electrodes at 142 Hz.
• Employed Random Forest and Convolutional Neural Network classifiers on spectral, temporal, and frequency domain features.
• Compared plant recordings near two performers executing gestures versus isolated control plants.

Main Results:

• Random Forest achieved 62.7% accuracy in distinguishing human movement proximity from control, a significant improvement over chance.
• Individual performer signatures were detectable (68.2% accuracy), while plant species had minimal impact (44.5% accuracy).
• Plants exposed repeatedly showed less negative bioelectric amplitudes compared to single-exposure plants.

Conclusions:

• Plants exhibit modest, statistically detectable bioelectric differences when humans move within approximately 1 meter.
• The study demonstrates a correlation between human proximity and plant bioelectric signals, not a causal mechanism.
• Underlying biophysical pathways and contributing factors like airflow or electromagnetic fields require further investigation.