http://dissertations.umu.ac.ug/xmlui/handle/123456789/18 Tue, 07 Apr 2026 14:40:58 GMT 2026-04-07T14:40:58Z http://dissertations.umu.ac.ug/xmlui/handle/123456789/1789 An enhanced hybrid model for credit scoring using machine learning approach Mugumya, Micheal This study investigates the application of a hybrid machine learning model for classifying loan statuses, combining logistic regression and decision tree classifiers. The dataset used comprises various loan records, with the target variable being the loan status, which is dichotomous in nature. The primary objective of this research is to develop a robust predictive model that can accurately determine the likelihood of a loan default. The analysis began with data preprocessing, including the handling of missing values and encoding of categorical variables. The dataset was then divided into training and testing sets to evaluate model performance. Two individual models’ logistic regression and decision tree were initialized with class weighting to address potential class imbalances. These models were combined using a soft voting classifier to form a hybrid model, leveraging the strengths of both algorithms. The hybrid model was trained and tested, with its performance evaluated using key metrics such as accuracy, precision, recall, F1-score, and confusion matrix. The results indicated that the model achieved a reasonable level of accuracy, particularly in predicting non-default loans (class 0), as evidenced by a high number of true negatives. However, the model's performance in predicting default loans (class 1) was less satisfactory, with a notable number of false negatives, suggesting a need for further refinement. Visualizations, including the confusion matrix and bar plots of evaluation metrics, provided deeper insights into the model's predictive capabilities and highlighted areas where the model could be improved. These findings underscore the complexity of loan status prediction and the challenges associated with imbalanced datasets. Overall, this study demonstrates the potential of hybrid machine learning models in financial risk prediction, while also identifying critical areas for future research and model enhancement. The implications of this research extend to financial institutions seeking to improve their risk management practices and enhance the accuracy of their loan approval processes. Brain Kasozi; Brain Kasozi Sun, 01 Sep 2024 00:00:00 GMT http://dissertations.umu.ac.ug/xmlui/handle/123456789/1789 2024-09-01T00:00:00Z http://dissertations.umu.ac.ug/xmlui/handle/123456789/1782 Teachers’ competence in the use of ICT and students’ academic performance in ICT Anatoli, Kirigwaijo Uganda like other developing countries is still in the initial stages of integrating ICT in teaching-learning process (Nyakito et al., 2021). The effective integration of ICT into classroom practices poses a challenge to different stake holders in the education sector, yet the use of ICT is essential in improving the learning abilities of learners. The aim of this study was to examine the impact of teachers‟ competence in the use of ICT to the academic performance of students in ICT in secondary schools in Hoima city. The research employed a mixed method study in which the target population consisted of two secondary Schools purposively considered due to adequate information regarding the variables of this study. Head teachers, teachers, and students of the two selected secondary schools in Hoima city were the unit of inquiry because they were deemed to be having accurate information about the study. The study used a size of 281 Respondents including; head teachers, teachers, computer laboratory assistants and students who were sampled using the Krejcie and Morgan (1970) approach. Primary data was collected using survey, interview, and documentary review methods. Self-administered questionnaires and Interview guide for teaching, nonteaching staff and students were used to obtain data as well. Findings revealed that teachers generally show strong competence in simpler ICT tasks such as opening programs and transferring files. However, they struggle significantly with tasks like searching the internet, troubleshooting, and connecting computers to projectors. The results varied across the attributes of academic performance and showed that majority of schools (65%) have a fair performance in ICT, with very few schools (1.9%) achieving a very good level of performance. In other findings, the level of teachers‟ compliance to the policy of using ICT in teaching and learning was generally poor, with majority of respondents (53.8%) indicating that teachers poorly comply with bringing ICT tools to their lessons. Determinately, this study highlights that lack of the necessary ICT skills, low compliance with ICT policies in teaching as well as the current state of ICT training being largely the multifaceted issues rooted in several systemic and individual factors hindering students‟ performance in ICT within secondary schools in Hoima city. Therefore, enhancement of teachers‟ competence by the appropriate stakeholders through additional training in ICT is highly recommended. This will subsequently translate into improving teacher‟s skills in ICT use, hence student‟s academic performance in ICT. Mark Kiiza; Mark Kiiza Sun, 01 Sep 2024 00:00:00 GMT http://dissertations.umu.ac.ug/xmlui/handle/123456789/1782 2024-09-01T00:00:00Z http://dissertations.umu.ac.ug/xmlui/handle/123456789/1780 Participatory action design science framework for adoptable solid waste management information systems Mark, Kaganda Jude Solid waste management is defined as the discipline associated with control of generation, storage, collection, transportation or transfer, processing and disposal of solid waste materials in a way that best addresses the range of public health, conservation, economic, aesthetic, engineering, and other environmental considerations (Rick et al, 2020). Urban informatics is an interdisciplinary approach to understanding, managing and designing the city using systematic theories and methods based on new information technologies (Wenzhong et all, 2021). Urban planners have adopted urban informatics research to help solve the problem and many mobile applications have been developed but not adopted owing to the use of a predictive System Development Lifecycle (SDLC) approach which excludes active involvement of the local population in the stages of SDLC and thus, not solving the challenge of information flow in solid waste management. The major objective of this research is to propose an adaptive SDLC approach and that is, using Participatory Action Design Science Research methodology (PADRE), for studies in the Urban Informatics domain and in this case dealing with solid waste management. A solid waste collection management system development conceptual framework is designed as proof of concept to show the strength of PADRE in developing new technological means to resolve contemporary issues or support everyday life in urban environments. This research further gives a comparison between using predictive SDLC approach and an adaptive SDLC approach in developing urban systems. The success in solid waste management in cities requires collaborative approaches of communities, NGOs, CBOs, Private institutions and Government in order to achieve the satisfaction level of solid waste management and make a clean environment. PADRE incorporates learning as an embedded nexus within each and every cycle. Hence, learning is an integrated component of each stage, and not a separate stage (Amir et al, 2018). Brian Kasozi; Brian Kasozi Sun, 01 Sep 2024 00:00:00 GMT http://dissertations.umu.ac.ug/xmlui/handle/123456789/1780 2024-09-01T00:00:00Z http://dissertations.umu.ac.ug/xmlui/handle/123456789/1775 A Robusta Coffee Leaf Image dataset for Improved Identification and differentiation Ssenoga, Gyaviira This study examines the possibility of designing and developing a classification model based on a dataset of Robusta Coffee seedling leaves taken with a Samsung Mobile phone camera. Images of leaves of various varieties (KR1-KR10) were taken in-situ and these were processed into a dataset that can be used in a machine learning pipeline to enable automation in Robusta Coffee variety identification and classification. Feature extraction is one of the major initial steps in any computer vision project. It deals with mining of important aspects from a population of features and rules out unnecessary detail that could become a burden in the machine learning process. Features are extracted from this dataset of coffee image leaves through Deep Learning (DL) and Convolutional Neural Networks (CNN) and using the python programming language and libraries such as Scikitlearn, Numpy, Matplotlib and Open Source Computer Vision (Open CV). Before feeding them into the pipeline, preprocessing has been carried out on each of the collected images to make it easier for processing. Due to computational limitations, only 350 images belonging to 5 classes namely; kr3, kr5, kr6, kr7 and kr9 were considered. These were run through data iterator that performed a series of augmentations to cater for data variability and randomness. The resultant model had an overall accuracy of 6.89% and a validation loss of 0.43%. Here is the link to the notebook https://github.com/GyaviiraS/KR-Classification- Model/blob/main/classification%20model%20100%20(1).ipynb It was discovered that most existing automatic computer vision systems have been designed in laboratory conditions on data captured with high end gadgets that cannot be accessed by end-users and that these solutions have not trickled down to the end users due to their complexity and cost of implementation. It has been demonstrated in this study that mobile phones can be used to create a more realistic dataset for machine learning and the solution created out of this dataset can be implemented by end user Godfrey Kagezi; Godfrey Kagezi Sun, 01 Sep 2024 00:00:00 GMT http://dissertations.umu.ac.ug/xmlui/handle/123456789/1775 2024-09-01T00:00:00Z