Published in Scientific Papers. Series E. Land Reclamation, Earth Observation & Surveying, Environmental Engineering, Vol. XII
Written by Stefan-Mihai PETREA, Ira-Adeline SIMIONOV, Alina ANTACHE, Aurelia NICA, Cristina ANTOHI, Dragos Sebastian CRISTEA, Adrian ROȘU, Valentina CALMUC, Bogdan ROȘU
The present study results are based on the application of XGBoost machine-learning algorithms and indicate that total waste, as a dependent parameter, can be accurately evaluated considering plastic wastes (feature importance-FI = 1.53, Rsq.= 0.75, RMSE = 0.47) in the case of V4 group, while for Romania, the dependent parameters identified as most reliable are chemical wastes (FI = 0.58) and industrial effluent sludges (FI = 0.04), with lower accuracy metrics (Rsq. = 0.46, RMSE = 0.75). In terms of waste treatment (WT), the portable batteries and accumulators’ market (FI=0.45) presents high reliability to be used as the main predictor (Rsq. = 0.80, RMSE=0.42) for V4 support tool, while for Romania, the waste generation (FI = 1.57, Rsq.= 0.85, RMSE=0.36) highly explains the variability of WT. However, batteries and accumulators waste (FI = 0.77, Rsq. = 0.82, RMSE=0.39) can be used as a reliable predictor for WT variation in a more extended analytical framework, in the case of Romania. It can be concluded that waste decision support management can be supported based on ML models which are different in the case of Romania, compared to V4, emphasizing the regional importance when developing environmental modeling-based tools.
[Read full article] [Citation]
