JavaScript is disabled for your browser. Some features of this site may not work without it.
| dc.contributor.author | BIZUNEH, MEKONEN | |
| dc.date.accessioned | 2023-06-16T07:18:51Z | |
| dc.date.available | 2023-06-16T07:18:51Z | |
| dc.date.issued | 2023-03 | |
| dc.identifier.uri | http://ir.bdu.edu.et/handle/123456789/15375 | |
| dc.description.abstract | Now adays, a variety of models have been developed and applied for groundwater table depth forecasting. These models can be categorized into the empirical time series model, physical descriptive model and machine learning models. The major advantages of the machine learning models is not data intensive than empirical approach and physical descriptive models pertaining to aquifer, geology and soil properties. It needs only Groundwater level (GWL) data and its explanatory features. However, in Ethiopia, it has not been applied previously to predict groundwater levels spatially. Therefore, this study aimed to evaluate different machine learning models for groundwater level prediction spatially and to prepare a shiny web app for the user by comparing Random Forest (RF), Decision Tree (DT), Support Vector Regression (SVR) and Artificial Neural Network (ANN) models. To predict the groundwater level, eleven input parameters were used such as; Normalized Vegetation Index (NDVI), Land Surface Temperature (LST), Topography Witness Index (TWI), Slope, Distance to Perennial River (DPR), Lineament Density (LD), Drainage Density (DD), Elevation (DEM) and Short Wave Transformed Reflectance (STR) obtained from remote sensing data, Rainfall obtained from National Meteorology Agency and Groundwater level(GWL) obtained from field data for 122 observation wells. Based on the RF model variable importance weight output the Elevation, LD, RF, Slope and TWI were of higher importance than other factors. The proposed machine learning model was built in R-studio using different packages by writing the code and trained it using 90% of the data. It was also tested using 10% of the data. Furthermore, the model’s performance was evaluated by relative root mean square error (RMSE), mean square error (MSE) and coefficient of determination (R2). During the training, the ANN model showed the best performance than the others which is 0.5, 0.25, and 0.98 values of RMSE, MSE and R2 respectively. Whereas, the RF model showed the best performance in the testing phase which were 1.57, 2.47, and 0.82 values of RMSE, MSE and R2 respectively. The data confirmed that the RF was the best machine learning model relative to the other models to predict groundwater level spatially due to its performance was high in the testing phase . Hence, the Shiny web app for the user was prepared using this model on the server side. Finally; this research recommended collecting more observation well data and accurate measurement of groundwater levels to observe the best machine learning model performance. Key words; ANN, DT, RF, SVR, Shiny Web App, Groundwater level (GWL | en_US |
| dc.language.iso | en_US | en_US |
| dc.subject | Civil and Water Resources Engineering | en_US |
| dc.title | EVALUATION OF VARIOUS MACHINE LEARNING MODELS AND DEVELOPING A WEB APP TO PREDICT DAILY GROUNDWATER LEVEL SPATIALLY IN GILGLE ABAY WATERSHED | en_US |
| dc.type | Thesis | en_US |
