Discussions

• All three models performed reasonably well even with these small datasets and few features.
• Predicting negative concentration was a problem found in many relevant studies. But none of the models tested in our work predicted any negative concentration.
• Overall, Neural Network showed the best accuracy and k-NN was the simplest and easiest.
• Random Forest showed better fitting abilities for higher concentrations than Neural Network.
• The performance of both Random Forest and Neural Network will be even better if a bigger dataset is supplied. Using more input features can also improve model’s performance.