Lai, Lik Sheng (2020) Feasibility Of Bootstrap Aggregating Fusion Method To Enhance Extreme Learning Machine For Reference Evapotranspiration Estimation. Final Year Project, UTAR.
Abstract
Evapotranspiration (ET) is a process comprising of both evaporation and transpiration, which plays an important role in the hydrological cycle. A good precise estimation of it is very important in various fields including water resources, agriculture and irrigation systems. The purpose of the study is to estimate the reference evapotranspiration (ETo) in Peninsular Malaysia using extreme learning machine (ELM) and the ELM enhanced with the bootstrap aggregating fusion method, with climatic data as input to the model. The climatic data used to train the model included maximum temperature, mean temperature, minimum temperature, relative humidity, wind speed, and solar radiation. These data were obtained from eight stations in Peninsular Malaysia, which were the Alor Setar, Bayan Lepas, Ipoh, Kuala Lumpur International Airport (KLIA) Sepang, Lubok Merbau, Pulau Langkawi, Sitiawan and Subang stations. The data obtained were arranged into 63 combinations and each of these combination sets was used separately as input in the model estimation. The results generated were interpreted based on the root mean square error (RMSE), Nash-Sutcliffe model efficiency coefficient (NSE), adjusted NashSutcliffe model efficiency coefficient (ANSE), mean bias error (MBE) as well as the mean absolute error (MAE). The results showed the best performance for most of the stations was the combination set with six climatic data as input for the model. Solar radiation was found to be the most important single input data for good model estimation. Bootstrap aggregating, also known as bagging did not improve but had reduced the performance of the model. A large amount of dataset utilized might be the reason for the inability of bagging to improve the performance of the model. Bootstrapping a huge amount of dataset might lead to over-fitting and thus reduce the accuracy in return. The large data size with respect to the low data dimensionality might also contribute to the ineffectiveness of the bagging to improve the model prediction.
Actions (login required)
![[img]](http://eprints.utar.edu.my/style/images/fileicons/application_pdf.png)
