http://ir.bdu.edu.et/handle/123456789/10151 2001-01-13T05:52:23Z http://ir.bdu.edu.et/handle/123456789/16310 SENTIMENT ANALYSIS FOR CUSTOMER SATISFACTION: CASE OF ETHIOPIA ELECTRIC UTILITY SERVICE TEWODROS, MESERET MENGESHA Advances in internet and web technologies have brought diversified online platforms for people to engage in online chats, discussions, and access instant information delivered to their handheld devices. Social media platforms, such as Facebook and X, have played a paramount role by enabling people to share their ideas, emotions, feelings, or talk about anything that affects us one way or another. This study introduces a machine learning-based Amharic language Facebook sentiment classifier model designed to conduct sentiment analysis for the Ethiopian electric utility’s electric distribution service. In this sentiment classification experiment, one classical machine learning and three deep learning models were evaluated: Support Vector Machine (SVM), Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), and Bidirectional LSTM (Bi-LSTM). The performance evaluation of these models reveals distinct patterns in their accuracy rankings based on test outcomes. For multiclass classification, the SVM model achieved test accuracy of 74.8%, and in binary class classification, it demonstrated an impressive test accuracy of 89.18%. Following closely, the Bi-LSTM model emerged as a strong contender with test accuracy slightly lower at 73.55% in multi-class classification and a commendable test accuracy of 87% in binary class classification. The LSTM model showcased a test accuracy of 72.75% in multi-class classification and a commendable test accuracy of 87% in binary class classification. Lastly, the RNN model exhibited a test accuracy of 65.9% in multi-class classification, and in binary class classification, the test accuracy slightly dipped to 86.33%. 2024-02-01T00:00:00Z http://ir.bdu.edu.et/handle/123456789/16309 Amharic Text-to-Speech Synthesis Using Deep Learning Approach Tadele, Demeke Sntie Text-to-Speech (TTS) translation is a process that generates synthetic speech artificially for a variety of uses, including telephone services, reading electronic documents, and speaking models for handicapped people. Currently, many text-to-speech translation models are available for different languages such as English, Afan Oromo, Tigrigna, and Welaytta. However, research on the Amharic language is extremely rare, so the study suffers from some limitations. Speech is generated from natural language text using deep learning approaches. Standard and nonstandard words, such as numbers, abbreviations, money, and dates, both SWs and NSWs found in written texts in a language. These NSWs cannot be detected by an application of the "letter-to-sound" rule. In general, the previous work converted text to speech using a rule-based and Hidden Markov Model. The main problem of HMM-based synthesis is that certain features for speech synthesis are hard coded by humans, but they are not necessarily the best features to synthesize speech. Hence to solve this problem we used LSTM and BiLSTM deep learning approaches. Because Deep learning has the ability to learn complex patterns in data and synthesize speech not required hard coding by humans. The performance of the LSTM model of MCD, MSE, and MAE is 0.2961, 0.0940, and 0.2474 respectively. And The performance of the BiLSTM model of MCD, MSE, and MAE is 0.2910, 0.0916, and 0.2400 respectively. As we have computed the translation performance of these models the BiLSTM has better performance. The second performance measurement of subjective evaluation metrics MOS is used to measure the quality of ineligibility and naturalness 4.14 and 3.93 respectively. Keywords: Text-to-Speech translation, Deep learning, long short-term memory (LSTM), and Bidirectional LSTM. 2024-12-01T00:00:00Z http://ir.bdu.edu.et/handle/123456789/16308 SOLID WASTE CLASSIFICATION USING DEEP LEARNING AND IMAGE PROCESSING TECHNIQUES Ashagrie, Ayenew Wondifraw Deep learning is playing a significant role in computer vision, particularly in various application areas such as environmental sanitation, which relies on image-based information. The accumulation of solid waste materials in different regions can lead to health problems for both humans and animals. An alternate technique of waste segregation is required since manual waste segregation is a very risky, laborious, and time-consuming operation. The alternative strategy needs to be economical and timeefficient. Even a minor classification error can result in a number of significant environmental harms. To address such problems, the use of deep learning techniques in combination with digital images of waste materials is crucial. In the field of image-based applications, deep Convolutional Neural Network (CNN) architectures have gained recognition for their excellent performance in classifying solid waste images. In this study, the VGG16 and InceptionV3 neural architectures are proposed for the classification of solid waste materials using images obtained from Bahir Dar and its surrounding environment. The model uses RGB 224×224 pixel colored images as input and extracts features from them, resulting in a 1024-dimensional feature representation. The collected digital images of solid waste materials undergo preprocessing using different techniques, such as noise reduction with bilateral filter techniques and segmentation using region-based segmentation techniques. A comparison is made between two CNN architectures, namely inceptionV3 and VGG16, and the testing accuracy results in 82.01% accuracy for VGG16 and 76.47% accuracy for inceptionV3. Based on this evaluation, VGG16 is selected as the superior model. Keywords; Deep Learning, VGG16, Inception, Solid Waste 2024-03-01T00:00:00Z http://ir.bdu.edu.et/handle/123456789/16307 EFFICIENT LOAD BALANCING ALGORITHM FOR FOG COMPUTING USING THROTTLED LOAD BALANCING ALGORITHM MELAKU, BITEW The placement of resources and proper allocation of workload to respected resources is called load balancing. Load balancing confirms proper allocation of tasks. Several algorithms have been developed to satisfy the request of the client for the fog nodes. Consequently, the nodes will be dynamically configured by the fog computing platform and these nodes can be physically or virtually present in the computing environment. Therefore, by using an appropriate load balancing strategy, selecting the fog node must be planned properly. For the uniform distribution of incoming jobs between virtual fog nodes, the throttled load balancing approach is proposed in this work. Further the performance is evaluated or analyzed using cloudAnalyst simulator and compare its performance with existing round robin, particle swarm optimization and hill climbing algorithms. Simulation results have shown that the proposed algorithm has uniformly distributed the load between fog nodes. Key Words: Fog computing, fog node, Load Balancing, Virtual Machine, Response Time, CloudAnalyst, Resource Utilization. 2020-10-01T00:00:00Z