Results uncertainty of support vector machine and hybrid of wavelet transform‐support vector machine models for solid waste generation forecasting

Abstract

The prediction of municipal solid waste generation (MSWG) plays an important role in a solid waste management system. However, achieving the anticipated prediction accuracy with regard to the nonhomogeneous nature of waste and effect of various and out of control factors on MSWG is quite challenging. In this article, support vector machine (SVM), one of the artificial intelligence techniques, and hybrid of wavelet transform (WT) and support vector machine (WT‐SVM) are used to predict weekly time series of MSWG in Tehran and Mashhad cites during the period of January 2006–December 2011. To improve the performance of SVM model, considering the influence of noise and the disadvantages of traditional noise eliminating technologies, the wavelet denoising method is applied to reduce or eliminate the noise in MSWG time series. Since Data‐driven models such as SVM involve potential of uncertainty that is difficult to quantify, uncertainty determination is one of important gaps observed in SVM results analysis. Therefore, Monte Carlo method was used to analyze uncertainty of the model results. Results showed both models could precisely predict MSWG in Tehran and Mashhad cites. However, the preprocessing of input variables by WT led to develop a more accurate model for prediction of weekly MSWG in both cities. The uncertainty analysis also verified that the WT‐SVM model had more robustness than SVM and had a lower sensitivity to change of input variables. © 2013 American Institute of Chemical Engineers Environ Prog, 33: 220–228, 2014

A Multistep Chaotic Model for Municipal Solid Waste Generation Prediction

“…Based on this fact, we also include sARIMA as a contrast method in this article. As for nonlinear models, ANN (Zade and Noori, 2008;Noori et al, 2010) and SVM (Noori et al, 2009;Abbasi et al, 2012Abbasi et al, , 2013 are the most commonly used model for weekly MSW data. This article only chose two typical nonlinear ANN and SVM models (nonlinear autoregressive with exogenous input [NARX] and PLS-SVM) as contrast models because we focused on discussing the performance of the basic model structure and the preprocessing method (wavelet transform and principal component transform) proposed in these articles can be implemented on chaotic model in further discussion.…”

“…Specifically, prediction of municipal solid waste (MSW) generation plays an important role in planning and management of urban solid waste system as prediction results can provide basic information for improving the effectiveness and efficiency of the planning system (Beigl et al, 2008;Chung, 2010;Lebersorger and Beigl, 2011;Abbasi et al, 2013). Nowadays, rapid urbanization has caused issues, such as population fluctuation, employment changes, and economic development (Dyson and Chang, 2005).…”

“…Correlation, bivariate regression analysis, group comparison, and time series analysis only need to consider one independent variable to build the models, while methods, such as multiple regression analysis, system dynamics, and input-output analyses, consider the complicated interactions between multiple variables, making these models far more complex and difficult to validate. Some of the latest intelligent models, such as artificial neural network (ANN) (Noori et al, 2010), partial least square support vector machine (PLS-SVM) (Abbasi et al, 2012(Abbasi et al, , 2013, grey system theory, and fuzzy dynamic models (Chen and Chang, 2000), have been proved to have good prediction performances for weekly or monthly time series data. In addition, various prediction models, including seasonal autoregressive integrated moving average (sARIMA) model and fuzzy logic model, have been developed to forecast daily MSW generation (Navarro-Esbrı et al, 2002;Karadimas, et al, 2006).…”

Prediction of Municipal Solid Waste Generation for Developing Countries in Temporal Scale: A Fuzzy Inference System Approach

“…Of the artificial intelligence models, ANN models have been found to predict MSW generation more accurately than conventional regression analysis and time series analysis (Inthariat et al, 2015). ANN model also requires large historical data along with the problems of over-fitting in training of the models, difficulty in the determination of network architecture (Abbasi et al, 2014). Similarly, it cannot handle linguistic data.…”

“…In artificial intelligence models, the relation between input and output variables are first established from previous experiences and then future outputs will be anticipated (Abbasi et al, 2014). Without the need of complete discernment of MSW generation process, these datadriven models have high ability to model waste generation fluctuations in temporal scale.…”