الفهرس 
chapter 1Only 14 pages are availabe for public view
 |  | from | 97 |  |  |
| | | from | 97 | | |
Abstract
DataStream prediction is a complex task in Machine Learning (ML), especially when confronted with continuous and large-scale data. Traditional regression and classification models struggle to handle the dynamic nature of streaming data and maintain real-time prediction accuracy. This Thesis proposes a framework for DataStream regression that employs a Multi-Processor Long Short-Term Memory (MPLSTM) architecture. The MPLSTM framework capitalizes on the inherent parallelism of Long Short-Term Memory (LSTM) networks, allowing for efficient and scalable streaming data processing. By dividing the DataStream into multiple parallel sequences, each processed by a parallel LSTM model, the framework achieves both high prediction accuracy and computational efficiency. The proposed framework addresses the challenges DataStream regression poses by effectively capturing the temporal dependencies and long-term patterns inherent in streaming data. The MPLSTM model adeptly learns and adapts to evolving data distributions, ensuring accurate predictions. It utilizes the power of parallel processing to handle the continuous influx of streaming data, enabling real-time prediction capabilities. The parallel LSTM models leverage the architectural design of MPLSTM to process multiple parallel sequences simultaneously, accelerating the prediction process without sacrificing accuracy. To evaluate the performance of the MPLSTM framework, extensive experimental evaluations are conducted on real-world datasets. These evaluations demonstrate the superiority of the MPLSTM framework over previous approaches, with an average prediction accuracy improvement of 15% and a reduction in computational time by 25%. The framework consistently achieves higher prediction accuracy and computational efficiency, showcasing its effectiveness in handling the challenges of DataStream regression. The results of the experiments reveal the capability of the MPLSTM framework to effectively capture the dynamic patterns and complex dependencies present in streaming data. The framework’s ability to adapt to evolving data distributions and maintain accurate predictions is an asset for real-time applications. The experiments also highlight the scalability and versatility of the MPLSTM framework, making it suitable for a wide range of DataStream regression tasks.