Abstract: Plant sensitivity and its bio-effects on non-thermal weak radio-frequency electromagnetic fields (RF-EMF) identifying key parameters that affect plant sensitivity that can change/unchange by using big data analytics and machine learning concepts are quite significant. Despite its benefits, there is no single study that adequately covers machine learning concept in Bioelectromagnetics domain yet. This study aims to demonstrate the usefulness of Machine Learning algorithms for predicting the possible damages of electromagnetic radiations from mobile phones and base station on plants and consequently, develops a prediction model of plant sensitivity to RF-EMF. We used rawdata of plant exposure from our previous review study (extracted data from 45 peer-reviewed scientific publications published between 1996-2016 with 169 experimental case studies carried out in the scientific literature) that predicts the potential effects of RF-EMF on plants. We also used values of six different attributes or parameters for this study: frequency, specific absorption rate (SAR), power flux density, electric field strength, exposure time and plant type (species). The results demonstrated that the adaptation of machine learning algorithms (classification and clustering) to predict 1) what conditions will RF-EMF exposure to a plant of a given species may not produce an effect; 2) what frequency and electric field strength values are safer; and 3) which plant species are affected by RF-EMF. Moreover, this paper also illustrates the development of optimal attribute selection protocol to identify key parameters that are highly significant when designing the in-vitro practical standardized experimental protocols. Our analysis also illustrates that Random Forest classification algorithm outperforms with highest classification accuracy by 95.26% (0.084 error) with only 4% of fluctuation among algorithm measured. The results clearly show that using K-Means clustering algorithm, demonstrated that the Pea, Mungbean and Duckweeds plants are more sensitive to RF-EMF (p <= 0.0001). The sample size of reported 169 experimental case studies, perhaps low significant in a statistical sense, nonetheless, this analysis still provides useful insight of exploiting Machine Learning in Bioelectromagnetics domain. As a direct outcome of this research, more efficient RF-EMF exposure prediction tools can be developed to improve the quality of epidemiological studies and the long-term experiments using whole organisms.

Malka N. Halgamuge, “Machine Learning for Bioelectromagnetics: Prediction Model using Data of Weak Radiofrequency Radiation Effect on Plants” International Journal of Advanced Computer Science and Applications(IJACSA), 8(11), 2017. http://dx.doi.org/10.14569/IJACSA.2017.081130

@article{Halgamuge2017,
title = {Machine Learning for Bioelectromagnetics: Prediction Model using Data of Weak Radiofrequency Radiation Effect on Plants},
journal = {International Journal of Advanced Computer Science and Applications},
doi = {10.14569/IJACSA.2017.081130},
url = {http://dx.doi.org/10.14569/IJACSA.2017.081130},
year = {2017},
publisher = {The Science and Information Organization},
volume = {8},
number = {11},
author = {Malka N. Halgamuge}
}