Predicting Wheat Yield from Time Series NDVI Values Using A Hybrid Arima-Bilstm Model With Attention Mechanism
| dc.contributor.advisor | Fitsum Assamnew (PhD) | |
| dc.contributor.author | Hiwot Teshome | |
| dc.date.accessioned | 2025-10-07T09:14:33Z | |
| dc.date.available | 2025-10-07T09:14:33Z | |
| dc.date.issued | 2025-06 | |
| dc.description.abstract | Two additional model approaches are also discussed in our work one being a Window Sliding approach where BILSTM Model is applied right after the failure point for ARIMA model is known by setting the window size(rolling window) and performing dynamic threshold failure detection using ARIMA validation RMSE metric result as a reference.Any value that is twice as much as the RMSE validation result is taken as a failure point. While the other model approach is a BILSTM model with an attention mechanism added to it. The HYBRID model showed an excellent performance on all metric results by having a lower result in comparison to metrics of ARIMA when it resulted in 98.21 percent on MSE ,a 89.43 percent on MAE, a 86.63 percent on RMSE and a 89.94 percent on the MAPE lower in comparison to its counterpart. In contrary the BILSTM model without attention mechanism which showed lower metric results than the HYBRID when it resulted in 72.54 percent on MSE,a 36.97 percent on MAE, a 31.35 percent on RMSE and a 23.72 percent on MAPE. During performance comparison of the HYBRID model against the Window Sliding model it was found that the HYBRID achieved a lower error value on all the metrics by having a 61.15 percent on MSE, a 37.27 percent on MAPE,a 37.67 percent on RMSE and a 30.41 percent on MAE .In addition we also found the failure point for the ARIMA model is at the middle of the year of 2016 for the Window Sliding model. | |
| dc.identifier.uri | https://etd.aau.edu.et/handle/123456789/7466 | |
| dc.language.iso | en_US | |
| dc.publisher | Addis Ababa University | |
| dc.subject | Normalized Difference Vegetative Index ( NDVI) | |
| dc.subject | ARIMA-BiLSTM | |
| dc.title | Predicting Wheat Yield from Time Series NDVI Values Using A Hybrid Arima-Bilstm Model With Attention Mechanism | |
| dc.type | Thesis |