APPLICATION OF SUPPORT VECTOR MACHINE AND NEURAL NETWORK TO FORECASTING CONSTRUCTION QUANTITY IN THAILAND

Abstract

The objective of this research is to study the application of support vector machine techniques and neural network techniques for forecasting construction quantity in Thailand. The selection factors used in the development of support vector machine and neural networks models to predict other construction volumes the government in Thailand consists of 4 factors: T35, Disposal of Current Revenue, T72, Private, T109, Non-residential and T122, Total C.i.f. Value, a forecasting model using vector machine support techniques. Uses the Radial Basis Function (RBF) kernel function to map the data. The predictive model using the neural network technique used the Gaussian function to convert values, used the cross-validation method, divided the data into 10 groups (10-fold cross validation) Teaching and testing data to measure model performance. Model results are specified as Mean Absolute Percentage Error (MAPE). Techniques that support vector machine. The result for MAPE is 1.31 and the coefficient of determination (R²) is 0.998. and Techniques that Neural Network. The result of the MAPE is 6.41 coefficients. and the coefficient of determination, (R²) of 0.982 is closer to 1, indicating that the model is appropriate. Research results have clearly shown that a support vector machine modelling technique can be used. and the neural network in time series forecasting of construction volume compared with the construction estimate of the Office of the National Economic and Social Development Board found similar effects. This is useful for organizations as the forecasting process is relatively simple, essential for planning, decision-making and management.