Skip navigation
st. Mary's University Institutional Repository St. Mary's University Institutional Repository

Please use this identifier to cite or link to this item: http://hdl.handle.net/123456789/7884
Full metadata record
DC FieldValueLanguage
dc.contributor.authorAgize, Nigatu-
dc.date.accessioned2024-04-29T11:53:41Z-
dc.date.available2024-04-29T11:53:41Z-
dc.date.issued2024-02-
dc.identifier.urihttp://hdl.handle.net/123456789/7884-
dc.description.abstractWater is one of humanity’s most essential resources, and in order to ensure that the limited quantity of water is used effectively, water supply facilities are required. The global population is driving up both the demand for and consumption of water. Given the restricted amount of water resources on Earth, this presents a number of resource related difficulties. In order to tackle this issue, the paper uses past data to investigate and contrast seven recurrent neural network models for water use. The goal is to create a deep learning model that uses previous customer data to analyze and forecast Addis Ababa Water and Sewerage Authority's water usage. Data collection, preprocessing, feature extraction, hyperparameter tuning, model training and development, and performance evaluation are some of the rigorous experimental processes involved in our research. With the introduction of smart water meters, it is now possible to get information on residential water usage. The data from the city of Addis Ababa (Ethiopia) was used as a case study to manage its limited resources, being water supplies. However, it is essential to acknowledge persisting challenges, including issues related to model overfitting and the critical necessity for precise hyperparameter tuning. The result of this study presents the remarkable ability of water consumption prediction through applying deep learning models, such Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), bidirectional GRU (Bi-GRU), bidirectional LSTM (Bi-LSTM) and Attention based model. The performance of the model also evaluated using the three evaluation metrics of RMSE (Root Mean Squared Error), MSE (Mean Squared Error) and MAE (Mean Absolute Error). Thus, Bi-LSTM with Attention mechanism scores lowest loss value of prediction RMSE, MSE and MAE with values of 0.08, 0.0064 and 0.16 respectively. This implies that the model using the Attention mechanism performance better as compared to others. Therefore, Bi-LSTM with Attention is proposed for constructing water consumption prediction model for Addis Ababa Water and Sewerage Authority. We advise future research to incorporate huge dataset sizes with a greater variety and quantity of variables.en_US
dc.language.isoenen_US
dc.publisherSt. Mary's Universityen_US
dc.subjectWater Consumption prediction, Deep Learning, Machine Learning, Attention mechanismsen_US
dc.titleWater Consumption Analysis and Prediction Using Deep Learning Approachen_US
dc.typeThesisen_US
Appears in Collections:Master of computer science

Files in This Item:
File Description SizeFormat 
10. Nigatu Agize.pdf1.8 MBAdobe PDFView/Open
Show simple item record


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.