Comparison of Neural Network and Principal Component-Regression Analysis to Predict the Solid Waste Generation in Tehran

Abstract

Background: Municipal solid waste (MSW) is the natural result of human activities. MSW generation modeling is of prime im­portance in designing and programming municipal solid waste management system. This study tests the short-term pre­diction of waste generation by artificial neural network (ANN) and principal component-regression analysis. Methods: Two forecasting techniques are presented in this paper for prediction of waste generation (WG). One of them, multivari­ate linear regression (MLR), is based on principal component analysis (PCA). The other technique is ANN model. For ANN, a feed-forward multi-layer perceptron was considered the best choice for this study. However, in this research af­ter removing the problem of multicolinearity of independent variables by PCA, an appropriate model (PCA-MLR) was de­veloped for predicting WG. Results: Correlation coefficient (R) and average absolute relative error (AARE) in ANN model obtained as equal to 0.837 and 4.4% respectively. In comparison whit PCA-MLR model (R= 0.445, MARE= 6.6%), ANN model has a better results. How­ever, threshold statistic error is done for the both models in the testing stage that the maximum absolute relative error (ARE) for 50% of prediction is 3.7% in ANN model but it is 6.2% for PCA-MLR model. Also we can say that the maxi­mum ARE for 90% of prediction in testing step of ANN model is about 8.6% but it is 10.5% for PCA-MLR model. Conclusion: The ANN model has better results in comparison with the PCA-MLR model therefore this model is selected for prediction of WG in Tehran.

Keywords

• Prediction of waste generation
• Artificial neural network
• Multivariable linear regression
• Principle component analysis