Master of Science/Engineering in Computer Science and Engineering

Browse

Now showing 1 - 3 of 3

A deep dive into node-level analysis with fusion RNN model for smart LTE network monitoring
(BRAC University, 2023-09) Islam, Md Rashidul; Alam, Golam Rabiul
Predicting and understanding traffic patterns have become important objectives for maintaining the Quality of Service (QoS) standard in network management. This change stems from analyzing the data usage on cellular internet networks. Cellular network optimiser frequently employ a variety of data traffic prediction algorithms for this reason. Traditional traffic projections are often made at the high-level or generously large regional cluster level and therefore has the lacking in precised forecation. Furthermore, it is difficult to obtain information on eNodeB-level utilisation with regard to traffic predictions. As a result, using the conventional approach causes user experience degradation or unnecessary network expansion. Developing a traffic forecasting model with the aid of multivariate feature inputs and deep learning techniques was one of the objective of this research. It deals with extensive 6.2 million real network time series LTE data traffic and other associated characteristics, including eNodeB-wise PRB utilisation. A cutting-edge fusion model based on Deep Learning algorithms is suggested. Long Short-Term Memory (LSTM), Bidirectional LSTM (BiLSTM), and Gated Recurrent Unit (GRU) are three deep learning algorithms that when combined allow for eNodeB-level traffic forecasting and eNodeB-wise anticipated PRB utilisation.The proposed fusion model’s R2 score is 0.8034, outperforms the conventional state-if-the-art models. This study also proposed a unique method that thoroughly examines individual nodes for the Smart Network Monitor. This approach follows adjustments made to soft capacity parameters at the eNodeB level, aiming for immediate improvement or long-term network growth to meet a consistent QoS standard. The algorithm relies on expected PRB utilization.
Empowering mobile network planning through deep learning: a path to democratization
(BRAC University, 2023-09) Nabi, Syed Tauhidun; Alam, Golam Rabiul
In the realm of cellular network internet data traffic assessment, the imperative task of forecasting and comprehending traffic patterns assumes pivotal significance for the effective management of network-designed Quality of Service (QoS) benchmarks. Conventional methodologies employed for predicting data traffic often suffer from inaccuracies. These traditional traffic forecasts, typically conducted at a higherlevel or within generously sized regional cluster contexts, tend to exhibit limitations in terms of accuracy. Furthermore, the absence of readily accessible eNodeB-level utilization data in conjunction with traffic forecasting exacerbates these challenges. This, in turn, may lead to compromised user experiences or unwarranted network expansion decisions based on outdated methodologies. This research embarks upon an ambitious journey encompassing an extensive dataset encompassing 6.2 million real network time series data points derived from Long-Term Evolution (LTE) networks. It also delves into associated parameters, including eNodeB-wise Physical Resource Block (PRB) utilization. The core objective revolves around the development of a traffic forecasting model that harnesses multivariate feature inputs and cutting-edge deep learning algorithms. Various advanced deep learning algorithms, including Long Short-Term Memory (LSTM), Bidirectional LSTM (BiLSTM), and Gated Recurrent Unit (GRU), have been separately tested for training purposes, with the most suitable model being chosen among the three for eNodeB-level predictions. This state-of-the-art deep learning model not only enables highly granular eNodeB-level traffic forecasting but also provides insights into anticipated eNodeBwise PRB utilization. The selected optimal deep learning model, BiLSTM, achieves a robust R2 score of 0.793, notably surpassing the performance of the other deep learning algorithms. Beyond the realm of PRB utilization, the study establishes a Quality of Service (QoS) threshold at 70% – a benchmark rooted in real network experience. This threshold serves as a pivotal trigger for decisions pertaining to soft parameter tuning. Leveraging the projected PRB utilization, the research introduces a pioneering algorithm designed to estimate eNodeB-level soft capacity parameter optimization. This algorithm empowers network operators to address short-term capacity enhancement solutions as well as long-term network expansion, all aimed at maintaining steadfast QoS benchmarks. Situated within the context of network planning, this study not only unravels the intricate dynamics of cellular data traffic but also catalyzes the concept of democratization. By harnessing the capabilities of deep learning, network operators are equipped with potent tools to navigate the intricate landscape of network optimization. Through this research endeavor, strides are made toward an envisioned future where technological advancements seamlessly converge with accessibility, thereby reshaping the contours of mobile network planning.
MEDNET – an approach to facial micro-emotion recognition using pixel binning and local Binary pattern - convolutional neural network
(BRAC University, 2023-09) Araf, Tashreef Abdullah; Alam, Golam Rabiul
Facial-Expression recognition is a very intriguing field of research, due to the complexity in its approach and applicability of widely available databases. However, Micro-expression recognition is quite a vague yet growing area of research due to its applicability in revealing minute facial expressions. These emotional triggers happen only under very pressing circumstances, which means detecting them can also be extremely tough due to shortage of time during which it lasts. In this study, the approach to Micro-facial expression detection is to explore passive and real-time observation that produces a great result for micro-facial expression recognition using a vast data set trained using new training techniques. A total of 59 papers were analyzed whose concepts were associative to our main thesis concept, which were categorized into three stages: Construction of a new dataset which constituted of standard and new facial images, which was trained using innovative image processing pipelines, implementation of a new Binary Pattern layer our Neural Network layer to accelerate the models expression tracking abilities, creation of a new facial model capable of facial and micro-facial expression recognition that performs better statistically when compared to its counterparts. Furthermore, the new model was tested in both artificial and real-world scenarios to accentuate the reliability of the data sources.