Abstract:
Municipal Solid Waste Management (MSWM) is one of the primary
tasks of metropolitan local authorities in developing countries. For
efficient and effective waste management schemes and scheduling,
accurate forecast of Municipal Solid Waste (MSW) generation is
essential, due to the uncertainties and unavailability of sufficient
MSW generation information and resources in developing
countries. The objectives of this paper are to identify influential
variables that affect the amount of MSW generation and to predict
the future MSW in Sri Lanka by consuming linear, nonlinear
models and machine learning technique and propose a model for
forecast future MSW generation using influential variables. Socio
economic data and waste generation data are collected from
Department of Census and Statistics and National Solid Waste
Management Support Center. Data preparation is done with
substitute missing values by average values. Pearson correlation
and Principal Component Analysis is used to find correlation
among influential variables. Linear model, Non-linear model and
machine learning model are used to forecast municipal solid waste
generation in Sri Lanka. Relatively Linear regression analysis,
artificial neural network (ANN) and Random forest used as linear
models, Non-linear model and machine learning model. Relatively
Correlation coefficient of linear regression classification, random
forest classification and ANN are R2=0.6973, R2=0.9608 and
R2=0.9923. Based on correlation coefficient, ANN provide higher
accurate result than linear regression and random forest models.
ANN is conducted by designing appropriate network architecture
with one neuron demand in the hidden layer. Based on the
analysed result a model was proposed to forecast future MSW
generation with four influential variables that are municipal solid
waste generation, total population, GDP growth rate, and Crude
birth rate.
