ICT
Permanent URI for this collectionhttp://10.0.100.25:4000/handle/123456789/32
Browse
Item Improved Enset Disease Detection with Image Processing and Convolutional neural network(2022) Tewokel AsratEnset (Ensete ventricosum Cheesman) is a perennial and herbaceous plant which widely distributed and cultivated in southern and southwestern Ethiopia as staple food for more than 20 million people in mixed subsistence farming systems. There are several diseases which tries to decline the yield with quality. However bacterial wilt and mealybug diseases of Enset are widespread in the country‟s major Enset growing regions causing losses of up to 100% destruction of farm fields in extreme cases. This paper looks into the use of deep learning to detect bacterial wilt disease and Enset mealybug, where data is obtained in small amounts and collected under minimally controlled conditions. Data augmentation is employed to get over the limits of the dataset sizeItem Design Of Active Control System Using Aco Based Pid Controller For Civil Structure Under Seismic Excitation(2023) Getachew LereboThe point of this research work is to develop structural control under seismic excitation. The background of this control mechanism is discussed in introduction part. In the second part, the reviews of literatures are discussed with their limitation and the third part includes the modeling of the civil building (2floor) structure to control, active control system with feedback mechanism has been developed. This feedback mechanism, PID controller with active mass damper is discussed with their designed specification and modeling. Active mass damper system and PID controller were designed for mitigating vibrations along the civil structure due to wind and earthquake. ACO and PSO algorithm are developed for tuning PID parameters. ACO based PID control mechanism has better performance and fast response. The sensor with dual properties has developed for vibration motion of the structure, this sensor is called piezoelectric sensor. Which can work with inverse operation and direct operation, in this thesis work direct operation or sensor operation was considered to sense vibration of the structure under earthquake. And 0.085ac voltage was produced. In spite of the ac output produced by piezoelectric element, rectifier is placed at the output side of the piezo element to convert ac to dc. And produced Dc voltage is low; boost converter is applied to step up this voltage and produces 11.18 Vdc. Dc motor as actuator was designed and produced mechanical force to apply in to active mass damper which is modelled in side of the civil structure. Actuator force produced is 1.9 ∗ 10−3N and disturbance force is −2.7 ∗ 10−5N.Item Classification Of Malt Barley Seed Using Ensemble Deep Learning Technique(2023) Dessie AbebeThe primary step in making beer is selecting the malt barley. Every malt house requires that the types of grain be checked before being purchased. Varietal uniformity is essential for the manufacture of high-quality malt. It is challenging to distinguish between different types of malt barley during inspection since it calls for training and experience. To tackle varietal selection difficulties Malt industries have employed and trained experts; even though, those experts do not work effectively due to tiredness, bias, and other factors. Therefore, many researchers are motivated to the development of classification model based on image processing to support experts across the world. Using digital image processing algorithms based on combined morphological, texture, and color features have been investigated to classify different varieties of Ethiopian malt barley.Item Model Development based on Demographic and Psychographic factor to improve performance in case of TVET using Machine Learning(2023) Habtom Atsebeha BerheTVET is important for the growth of human resources and, consequently, for the advancement and prosperity of a community. In this thesis, the factors that significantly affect TVET students' academic achievement and performance were modeled. The academic performance of TVET students has been found to be significantly influenced by a number of factors, including gender, age, monthly family income, study hours, stimulant use during the course of the study, and satisfaction with the area of study placement. According to this study, lower academic accomplishment was associated with a variety of factors, including having a negative opinion of TVET, being female, having a low family income, studying for a shorter amount of time, using stimulants while studying, and not being content with one's field of study.Item Detection and Classification of Skin Cancer Using Image Processing and Deep Learning(2023) Degarege Amare DemessieSkin cancer is one of the most serious malignancies today, and it kills so many individuals all over the world. It develops in the skin tissue and can harm nearby tissues, result in disability, or even result in death. Furthermore, in today‟s technologically advanced world, it is crucial that machines, not people to fix the issue. One of the finest approaches to address the issues with skin cancer is deep learning. A recent topic of research in contemporary technology that makes use of large data, virtual and augmented reality, and micro services is deep learningItem Classification Of Goat And Sheep Skin Diseases Using Deep Learning Approaches(2026) Hiluf Nuguse GideyAnimal farm has played one of critical roles in the socio economic wellbeing of most developing countries, Ethiopia being one of them as sheep and goats play major role in providing food security, generate incomes and export income. Nevertheless, there are a number of skin diseases, which have a severe impact on the quality, and yields of animal products. In the rural set ups, they are not handled well because of the inaccessibility of veterinary clinics and shortage of expertise in the field. In this paper, an enhanced method of deep learning approach provided to identify and classify skin diseases in sheep and goats with deep learning. The system automation CNN architectures mean that features are automatically trained by skin images by which skin diseases can be classified and classified to either bacteria and virus, parasite or healthy. The evaluated and tested model underwent testing and evaluation on the standard measures, which include accuracy, precision, recall and F1-score. The data employed in this thesis is 2341 images of data. The training dataset classified as 80% (1873 images), 10% (234 images) and 10% (234 images) in training, validation and test performance respectively.Item Predictive Analytics of Industry Skills Gaps: Using Machine Learning – In Case of Tilahun Yigzaw TVET College(2026) Azmach BerheAlthough Technical and Vocational Education and Training (TVET) have gained increasing significance, many graduates remain unable to secure employment due to a persistent mismatch between the skills they acquire and the requirements of the labor market. This issue continues to contribute to high levels of unemployment among trainees. Within this context, this study attempts to analyze Tilahun Yigzaw TVET College in Tigray, Ethiopia. The research concludes that predictive analytics have the potential to enhance data-driven decision-making in TVET institutions by identifying critical factors influencing the skills gap and facilitating the alignment of training programs with industry demands.Item Developing An Ai Integration Framework In Tvet Curriculum: The Case Of Aatpc(2026) Zerihun Basazin YehualaAI is increasingly transforming the education sector by enhancing personalized learning, automating, administrative task in teaching and learning process. However, the use of AI in TVET in Ethiopia is still in its initial stage with low infrastructure, low digital literacy among teachers and lack of a structured implementation strategy. This research attempts to fill these gaps by developing a contextualized framework for integration of AI into TVET curriculum using the case study of AATPC. The research use a mixed-methods research design. Focus groups and interviews were used to gather data from TVET instructors, administrators, curriculum developer. We also collected using questionnaire from students and instructors. Thematic analysis and descriptive statistics were used to analyze the data. The data was analyzed using thematic analysis and descriptive statistics. A total of 72 questionnaires were distributed. Out of these 69 were returned while 3 were not returned. Among the returned questionnaires 2 were excluded due to incomplete or invalid responses. Therefore, 67 valid questionnaires (93%) were considered valid for analysis. Validity and reliability tests were conducted to assess coverage of the topic, with a Cronbach’s alpha coefficient of 0.988.Item Design Double Layer Aerial With Lens For Millimeter Wave Image Detection Systems(2026) Ehtinat AmisaluThere has been significant interest in millimeter-wave systems used in detection of concealed images. security, medical and industrial fields because of their ability to detect concealed items with. minimal penetration loss. Current antenna designs, i.e. single-layer patch antennas. Slot designs, and APSAs have low gain, small bandwidth, low directivity, and are generally low gain. restrained detection range, and hence less image resolution and fidelity. To overcome these challenges, this work is a patch antenna of circular form on the double-layer with an inbuilt lens, optimized at 24.411 GHz - the best millimeter-wave radio frequency to use in imaging. The High Frequency Structure Simulator (HFSS) was used to develop and simulate design. utilizing multi-layer layouts and lenses in order to increase radiation effectiveness and beam. focusing. Performance tests were taken to be of importance with regard to gain, directivity, and reflection. coefficient (S11), voltage standing wave ratio (VSWR) and bandwidth which were subsequently correlated.Item Near-Real-Time Prediction of Desert Locust Presence Risk in Ethiopia Using Satellite-Derived Environmental Data and Machine Learning(2026) Belay AmareDesert locusts (Schistocerca gregaria) pose a long-standing challenge to crop production and food security in Ethiopia. Current warning systems have several limitations. These include relatively coarse spatial resolutions, rule-based inferences, and delayed reports. These limitations restrict their operability and effectiveness. This study developed a near-real-time, spatially validated framework for predicting locust presence in Ethiopia using multisource satellite data. The model integrates MODIS NDVI, CHIRPS rainfall, and ERA5 temperature datasets covering 2015–2024, combined with 18,742 quality-controlled locust occurrence records from the FAO DLIS and the DLCO-EA. All data were harmonized to a spatial resolution and a 16-day temporal interval. A total of 32 environmental predictors were derived, including lagged variables, anomalies, and seasonal indicators. Four models (LR, RF, XGBoost, and LightGBM) were tested using spatial block cross-validation. LightGBM achieved the best spatial performance (AUC = 0.740), and a stacked ensemble (LightGBM + XGBoost + calibrated logistic regression) achieved a final test set AUC of 0.780 (95% CI: 0.765–0.795) with a conservative temporal validation AUC of 0.85 under a strict two-year gap split.Item Automated Detection And Analysis Of Sentiment And Hate Speech In Himtagne Language Social Media Posts And Comments Using Deep Learning(2026) Ashagrie Goshu KassaHate speech on social media has become a critical global concern, particularly in under-resourced languages where automated moderation tools are limited or unavailable. In Ethiopia, while some progress has been made for languages such as Amharic, the Himtagne language remains largely unsupported, allowing harmful online content to spread unchecked. This study proposes an automated hate speech detection system for Himtagne social media text using deep learning techniques. A dataset of 9,258 Himtagne Facebook posts and comments was collected and manually annotated into five classes: non-hate, religious, gender-based, racial, and political hate speech. The study compares frequency-based feature extraction (TF-IDF) with embedding-based representation (FastText), and evaluates two classification models: Support Vector Machine (SVM) and Stacked Bidirectional Long Short-Term Memory (SBi-LSTM).Item A Machine Learning Model For Road Traffic Accident Factor Classification: The Case Of Kamba Woreda, Gamo Zone, Southern Ethiopia(2026) Amanuel Dulebo ShurgaRoad traffic accidents are events that occur on roads open to public traffic and result in injury, loss of life, vehicle damage, and substantial economic costs. As the number of vehicles increases worldwide, including in Ethiopia, road traffic accidents have risen accordingly, yet it remains difficult to determine the specific conditions that lead to them. Previous studies have largely focused on classifying accident severity levels or predicting whether an accident will occur, often using a limited set of features. This study addresses a different problem: classifying the contributing factors of road traffic accidents into four categories — human, vehicle, road, and environmental factors — using supervised machine learning. The dataset was obtained from the Gamo Zone Kamba Woreda Traffic Police Office and covers seven years of accident records (2011–2017 E.C.). After preprocessing, the dataset comprised 5,250 instances described by 19 attributes, including the target class.Item Machine Learning-Based Prediction Model of Obesity Risk Among adolescents in Central Ethiopia Regional State(2026) Abebash BeyeneThis study discovers the application of machine learning techniques to predict obesity risk among adolescents aged 10 - 19 years in Central Ethiopia Regional State. It uses a comprehensive dataset composed from above 10,000 adolescents. The methodology involves training multiple supervised machines learning classifiers such as Decision Trees, Support Vector Machines, Random Forest, and XGBoost using balanced and processed datasets. Metrics like as accuracy, precision, recall, and F1-score were used to assess the model's performance using 10-fold cross-validation. The findings demonstrate that Random Forest and XGBoost outperform other models with good predictive accuracy and stability, demonstrating their usefulness for classifying obesity in adolescents. The findings underscore the significance of combining behavioral, biological, and environmental data for efficient risk stratification and provide important insights into the predictive variables of teenage obesity in Ethiopia.Item IoT-Based Smart Waste Management System for Hawasa City(2026) Kehali Berihun AnagawThe problems of controlling municipal solid waste have increased because of global urbanization and population growth, creative and scalable solutions to rescue environmental sustainability. This research maintains a mechanism for creating effective, sustainable and intelligent urban systems with a collaboration of Internet of Things (IoT) technologies into garbage management. Having information from implementation models, case studies, and current research, this study offers a thorough design and implementation of IoT-based smart waste systems compatible for sustainable cities. Depending on the trash management optimization test, this research suggests the use of sensor-enabled waste cans, the real-time data analytics, and cloud-based controlling mechanisms to improve service responsiveness, boost resource allocation, and maximize collection efficiency. The case study in Ho Chi Minh City indicates how IoT-driven solutions can support with the transformation to a circular economy while addressing the traditional garbage management drawbacks. In addition data-informed policy planning, predictive maintenance and intelligent routing are made possible by the collaborations of machine learning and IoT. The study also encompasses at deployment issues in low-resource environments, like Malawi, highlighting the importance of technological design tailored to local context, community participation, and regulatory assistance.Item Predicting Of Stunting Among Under-Two Children Using Ensemble Machine Learning Approaches(2026) Weldebirhan Tadesse BerheChild stunting is defined as low height-for-age and is a major global public health issue affecting millions of children, mainly in Sub-Saharan Africa. Chronic malnutrition, exposure to conflict and drought, and poor health services are driving up the incidence of stunting in Ethiopia. The traditional methods based on retrospective surveys are not always timely enough and do not include multifactorial. To overcome these limitations, this study develops a predictable and interpretable ensemble machine learning model to identify under-two children at risk of stunting, so enable timely interventions and resource optimization. This research utilized Secondary source of data from the Ministry of Health (Seqota Declaration), including child-level attributes, maternal-level attributes, socioeconomic, nutritional, and environmental factors from the first 1,000 days of life. Five ensemble machine learning models, such as RF, Gradient Boosting, XGBoost, Light Boost, and CatBoost were combined with advanced feature selection (Boruta),Item Amharic Scanned Document Retrieval With Approach Of NLP(2026) Chernet Taye DareEthiopia possesses a vast and valuable collection of documents written in Amharic, ranging from administrative records and legal documents to historical and archival materials. In recent years, many of these documents have been digitized through scanning in an effort to preserve them and improve accessibility. However, despite digitization, effective access to the information contained in these scanned documents remains a major challenge.The primary difficulty arises from the limitations of existing Optical Character Recognition (OCR) technologies when applied to the Amharic script. Due to the visual complexity of the Ethiopic writing system, the morphological richness of the language, and variations in document quality, OCR systems often generate noisy and error-prone text. As a result, conventional information retrieval systems that rely on exact keyword matching perform poorly, making it difficult for users to locate relevant documents efficiently.This study addresses these challenges by proposing a rule-based Amharic scanned document retrieval system using Natural Language Processing techniques. The research follows a design science research approach, involving data acquisition from scanned Amharic documents, identification and formalization of linguistic and OCR-related rules, system design and implementation, and systematic evaluation using standard information retrieval metricsItem Ai Integrated Assistive Technology Framework To Enhance Learning Accessibility For Visually Impaired Students.(2026) Hailemariam Derese MekuriaInclusive Education is a foundation to most modern education programs that promotes equal learning opportunity access for all students, regardless of the differences they may have based on physical or cognitive characteristics to a higher quality education. This study explores the ability of assistive technology to read visual instructional content in education. Screen readers such as JAWS can read text, but cannot read diagrams, charts, and visual instructional content, such as diagrams, charts, and graphics, that is typically inaccessible to visually impaired Students. The study proposes JAIF through Design Science Research. It is based on a YOLOv12-based object detection model to recognize diagram elements, Optical Character Recognition (OCR) to extract text, Natural Language Processing (NLP) to generate semantic descriptions, and a Text-to-Speech module in JAWS that produces real-time audio output.Item An Explainable Ai Model For Early Identification Of At-Risk (Low Performance) Students In Ethiopian Tvt Colleges (The Case Of Alamata)(2026) Solomon Chekole TesfayEthiopian Technical Vocational Education and Training (TVT) institutions face persistent challenges regarding students’ academic performance. At Alamata TVT College, Certificate of Competency(CoC) examination failure rates exceeded 35% in 2018 and 31% in 2019. Despite the failure rates, the college has no predictive early warning system to identify at-risk trainees before these concerning figures, no early warning systems before they fail, leaving interventions reactive. This study develops and evaluates an explainable machine learning model to classify students into three performance risk categories: high, medium, and low risk, enabling targeted early intervention in Ethiopian TVT colleges.Item Amharic Text Chuncker Using Machine Learning(2026) Mesele Gawo EmoAmharic is one of the most widely spoken languages in Ethiopia and is commonly used in education, government offices, media and everyday communication. However, the development of Natural Language Processing (NLP) tools for Amharic is still limited mainly because of the lack of large, annotated datasets and effective language processing systems. One key NLP task is text chunking which means grouping words into meaningful parts such as noun phrases and verb phrases. This process is very important for many applications like machine translation, information retrieval, Question answering and text summarization. In this study an Amharic text chunking system is developed using machine learning techniques to overcome the limitations of previous rule-based and statistical approaches. The main goal is to design and evaluate models that can automatically identify phrase structures in Amharic text focusing on Conditional Random Fields (CRF) and Memory-Based Learning (MBL) methods using different features and context window sizes.Item Developing a Context-Aware IoT Framework for Traffic Management in Addis Ababa city.(2026) Abebaw IsraelUrban traffic congestion in Addis Ababa has become critical due to rapid population growth, increasing vehicle ownership, and outdated traffic control systems that rely on fixed timing signals and manual police intervention. This study addresses the lack of an intelligent, real time traffic management system tailored to the local context. The primary objective was to design and evaluate a Context Aware Internet of Things (IoT) framework that dynamically adjusts traffic signals and provides adaptive control based on real time contextual parameters such as traffic density, time of day, weather conditions, and emergency vehicle presence.
