Application of Artificial Intelligence on Hourly Weather Forecast to Improve the Accuracy of Short-Range Prediction in Ethiopia
| dc.contributor.advisor | Tibebu Kassawmar | |
| dc.contributor.advisor | Addisu Semie (PhD) | |
| dc.contributor.author | Atalel Wassie | |
| dc.date.accessioned | 2025-08-17T22:35:42Z | |
| dc.date.available | 2025-08-17T22:35:42Z | |
| dc.date.issued | 2023-09 | |
| dc.description.abstract | The importance of accurate weather forecast has grown in recent years, especially for those who rely on the state of the atmosphere and its phenomena in hourly, daily, monthly and yearly bases. The transportation (aviation), Agricultural activities, construction, recreational activities (sports, concerts) are particularly affected either hourly, daily or yearly bases. This study proposed a newly emerging approach to forecast the weather that can minimize economic and human losses caused by imprecision. Highly correlated input features were used for the study that was conducted on ten years of hourly data from 2013 to 2022 to forecast precipitation, temperature and fog. The proposed method used machine learning models for data preprocessing, while deep learning models are used to forecast targeted variables. Four deep learning models were built and evaluated accordingly. These were LSTM, BiLSTM, GRU, and Simple RNN. Three experiments were conducted with different hyperparameter configurations. Hence, the LSTM model was found to have the best performance comparatively, with the following error metrics. For the validation of temperature forecast: root mean square error (RMSE) of 0.136, mean squared error (MSE) of 0.018, and mean absolute error (MAE) of 0.111. The corresponding loss values for the training, testing, and validation sets were 0.018, 0.017, and 0.018, respectively. Similarly, a model also achieved RMSE = 0.023, MSE = 0.0005, and MAE = 0.008 of error metrics for the validation of precipitation. The losses for training, testing, and validation found to be 0.00094, 0.00094, and 0.00054, correspondingly. The LSTM model and the random forest classifier were compared for fog forecasting. The LSTM model achieved the following error metrics: RMSE = 0.036, MSE = 0.0013 and MAE = 0.0034. The loss values for training, testing, and validation were 0.0028, 0.0019, and 0.0013, individually. The random forest classifier achieved an accuracy of 99.96% for fog prediction. The outcomes of the study confirmed that the proposed method can be used to forecast the weather conditions accurately. This can help to minimize economic and human losses caused by forecast imprecision. | |
| dc.identifier.uri | https://etd.aau.edu.et/handle/123456789/6912 | |
| dc.language.iso | en_US | |
| dc.publisher | Addis Ababa University | |
| dc.subject | Artificial Intelligence | |
| dc.subject | BiLSTM | |
| dc.subject | Deep Learning | |
| dc.subject | GRU | |
| dc.subject | LSTM | |
| dc.subject | Loss | |
| dc.subject | Machine Learning | |
| dc.subject | Model | |
| dc.subject | Metrics | |
| dc.subject | Random Forest | |
| dc.subject | Simple RNN | |
| dc.subject | Weather Forecast | |
| dc.title | Application of Artificial Intelligence on Hourly Weather Forecast to Improve the Accuracy of Short-Range Prediction in Ethiopia | |
| dc.type | Thesis |