Speaker
Description
The study explored the potential of using Machine Learning (ML) techniques for the prediction of Polycyclic Aromatic Hydrocarbon (PAH) contaminants in diesel tainted soils. A bio-geotechnical method (i.e. phytoremediation) adjudged to be sustainable, low-energy soil remediation and stabilization techniques was employed. The plant (i.e. Megathyrsus maximus) proved to have the potential for use in phytoremediation with assistance from organic amendment using poultry manure. PAH contaminants were degraded and reduced to 3.77mg/kg from 47.25 mg/kg after 16weeks of field experiments. However, the deficiency of phytoremediation characterized with time consumption and tediousness, data mining techniques assisted in the forecast of the obliteration efficacy. Phytoremediation data/parameters generated from pot and field experiments (i.e. soil pH, organic carbon, and electrical conductivity, as well as the relative pollutant concentrations) were subjected to data mining categorization in machine learning techniques. Four ML algorithms (Random Forest, RF; Support Vector Machine, SVM; Multi-Layer Perceptron, MLP; and Linear Regression, LR) were deployed and compared to each other with the help of the 10-fold cross-validation tool of the machine learning software WEKA. Results from the evaluation parameters of the algorithms showed that RF algorithm which had a correlation coefficient of 0.9277 was better than the other algorithms (SVM = 0.8659, MLP = 0.872, and LR = 0.8749). Comparatively, the RF present the best algorithm in building a PAH prediction for diesel contaminated soils. As such the RF algorithm can be deployed for use as a learning tool in phytoremediation analysis.
Keywords: Bio-geotechnics, Megathyrsus maximus, poultry manure, PAH, Random Forest (RF).