Artificial neural network modelling of the amount of separately-collected household packaging waste. (10th February 2019)

• Record Type: Journal Article
• Authors: Oliveira, Verónica; Sousa, Vitor; Dias-Ferreira, Celia
• Is Part Of: Journal of cleaner production. Volume 210(2019)
• Page Start: 401
• Page End: 409
• Publication Date: 2019-02-10
• Subjects: Municipal solid waste -- ANN -- Genetic algorithm -- Regression model -- Urban waste -- Recycling; Factory and trade waste -- Management -- Periodicals; Manufactures -- Environmental aspects -- Periodicals; Déchets industriels -- Gestion -- Périodiques; Usines -- Aspect de l'environnement -- Périodiques; 628.5
• DOI: 10.1016/j.jclepro.2018.11.063
• Languages: English
• ISSNs: 0959-6526

Abstract

Abstract: This work develops an artificial neural network (ANN) model using genetic algorithms to estimate the annual amount (kg/inhabitant/year) of separately-collected household packaging waste. The ANN model comprises one input layer, one hidden layer with seven neurons and one output layer. Ten variables affecting the amount of separately-collected packaging waste were identified and used in the ANN model. These variables are related to the level of education of the population, the size and level of urbanisation of the municipality, social aspects related to poverty and economic power and factors intrinsic to the waste collection service. A comparison between ANN and regression models for the estimation of packaging waste is also carried out. The performance of the proposed ANN model for a data set of 42 municipalities located in the centre of Portugal, measured by the R 2, is 0.98. This value is 34% higher than the best regression model applied to the same data set (R 2 = 0.73), indicating that ANN has a significantly higher explanatory power than traditional regression techniques. Another advantage is that ANN is not as sensitive to outliers as regression. However, ANN is more complex, has a higher number of variables, and the model development and interpretation of the results are more difficult. Nevertheless, the higher performance of ANN makes it a valuable tool in the definition of strategies to increase recycling and achieve circular economy goals. Highlights: ANN technique is used to predict the amount of separately-collected packaging waste. ANN models performed better than regression models. The best performing ANN model achieved an R 2 of 0.98 The performance of ANN models was 34% higher than non-linear regression models. ANN technique is more difficult to model and interpret than regression.