MSIT Theses and Dissertations (2023)

Permanent URI for this collection

http://hdl.handle.net/11071/13518

Browse

Now showing 1 - 11 of 11

A Loan default prediction and loan amount recommendation tool for SACCOs in Nairobi: a case of Okoa Management SACCO
(Strathmore University, 2023) Mwalozi, P. M.
SACCOs loan delinquency is a severe danger to the organization's capacity to continue availing loans to loan applicants and to grow. SACCOs are unable to collect what they have lent out to loan beneficiaries as the default rate rises gradually. This research project aimed at using the analysis of the different factors that determine loan defaults in microfinance institu-tions, microlending institutions and SACCOs in Kenya with a focus on Okoa Management Ltd. and how the same factors can be used to predict the likelihood of a loan borrower to default in the repayment process by applying machine learning algorithms. Credit risk assessment pre-cision is important to the functioning of lending institutions. Traditional and most existing credit score models are developed and designed using demographic characteristics, historical payment data, credit bureau data and application data, with most of them not suitable for de-veloping countries such as Kenya which consider the employment type (casual, temporary, contractual or permanent) and the fact that we can lend up to 3 times as much as the borrower’s savings. With these factors being constantly changing and dynamic, credit risk models based on machine learning algorithms provide a higher level of accuracy in predicting default as they can be continuously trained with new data sets should the variables that are used change. Risk management has been an increasing issue for credit lending institutions as the need to deter-mine the likelihood of defaulting by borrowers is becoming more evident. By using machine learning, we can be able to reduce the uncertainty that comes with borrowing and even go further to recommending lower amounts for borrowers who we predict are likely to default in the repayment of the loan amount they have in mind. The research focused on three main al-gorithms: logistic regression, decision trees and tensor flow on the prediction. The algorithm that provided the best accuracy was the decision tree. The results of the research showed that people with little or no collateral (home-ownership/car ownership) were more likely to default and that there was a low correlation between months since last delinquent and the loan predic-tion default likelihood status. Keywords: Loan default prediction, machine learning, credit lending
A Model for sign language recognition for Kenyan sign language
(Strathmore University, 2023) Wanjala, G. K.
Computer vision aids in increasing tech accessible for communities that are underserved, such as the disabled community. This study demonstrates how artificial intelligence via computer vision helps to bridge the communication gap between those with hearing problems and the general population. The purpose of this paper is to bring forth an artificial intelligence solution to cater to this targeted group to aid in communication. Artificial intelligence has come a long way to solve this problem of enabling sign language notations to be translated into readable form that can be easily understood. This is in accordance with the fact that there is a collective duty to ensure that the deaf can be part of our society on an equal basis with others, free from discrimination even when it comes to speech and communication. There is a great need for this interpretation so that communication is sped up through translation. Understanding between the deaf and hearing people can be fostered as well as costs associated with training individuals in sign language communication in sign language training centers are minimized. To answer this question, the research work collected and analyzed photos and videos in a quasi-experiment consisting of target photos of Kenyan sign language notations. The model is trained on 9100 Kenyan sign language (KSL) notations of varied gestures spanning from health and wellness to common day to day basic notations such as greetings, expressing feelings among others. Transfer learning through Tensor Flow object detection model, Open CV framework for image processing and python was used to actualize this sign language translation model in this research work. A trained machine learning model organizes the input photos and videos, analyses them and produces text that maps to the corresponding sign language notation used. Individual users can use the model to translate Kenyan sign language notations into readable English text. The model gave performance levels of 85% accuracy on a 20,000 training steps for 40 epochs. This gave a perfect balance of training duration and accuracy levels on the dataset given. One of the notable findings was that notations that involved movement of the hands and other body parts to express gestures were harder to detect and translate due to the motions involved. A lot of training data on such notations is needed to train the model further in detecting them. Keywords: Artificial Intelligence, computer vision, machine learning, sign language, disability
A Smart water management system for detecting household water wastage
(Strathmore University, 2023) Gichuhi, J. M.
Water scarcity in Kenya is a significant issue that cannot be overlooked. Despite numerous efforts made by Water Service Providers in delivering water services to Kenyan residents, the high demand for accessible water across the country remains unmet. In addition, the provider’s ability to meet this demand is further impeded by various obstacles including inadequate control over water usage, insufficient resources to manage and conserve water, and the dilapidated water infrastructure within the Nation. As global water resources rapidly decline, due to the effects of climate change and overuse, it is crucial to take action in conserving this resource. This study looks into the development of a smart water management system using water flow sensors connected directly to the water appliances, Node MCU microcontrollers and a cloud-based application. The system focuses on monitoring household water usage frequency to inform users of their consumption and alert them via mobile notifications of potential water leaks to reduce wastage. Management and control of household water consumption is a positive step towards impacting water conservation efforts. Keywords: Internet of Things, Machine Learning, Usage Detection, Leakage Detection, Smart Water Management System
A Sound classification and display tool for assisting the deaf and hard-of-hearing: a case of Kenya
(Strathmore University, 2023) Wanjiru, R. W.
Sound is an essential component of existence in all aspects of life. It is a crucial component when it comes to creating automated systems for various domains such as personal safety and essential surveillance. Hearing people always absorb information through sound and language that is spoken around them. On the other hand, deaf and hard of hearing people lack the luxury of hearing and may end up having major problems due to lack of this awareness. Various researches have shown that there is a mismatch between the need and the demand of the assistive technologies as you find that the need is high but the demand and supply is low which impose a challenge in enhancing access of the assistive devices. Also, there is a gap between the number of people who require assistive technologies to meet their needs and the number of people who are willing and able to purchase and use these technologies. This mismatch could be due to factors such as the cost of the technologies, lack of awareness or knowledge about the technologies, or cultural barriers to their use. Only a small percentage of people have access to the assistive devices. This study reviewed the existing assistive technologies for the deaf and hard of hearing. Prior studies on assistive technologies for the deaf revealed that, sound classification systems have been developed world wide, but none have been implemented for use in Kenya. The research employed a machine learning approach, specifically utilizing convolutional neural networks, to design a sound classification model. The process involved transforming detected sound events into spectrogram images, which were then processed by the Convolutional Neural Network to extract relevant features. The extracted features were subsequently employed to classify environmental sounds, including car horns, dog barking among others. Once the sounds have been classified, a mobile app was used to display a notification indicating the type of sound that has been detected. The machine learning model was evaluated for its effectiveness in assisting the deaf and hard-of-hearing individuals, with the ability to accurately classify a wide range of urban sounds relevant to the study and display corresponding notifications on the user interface. The development of this model stems from a strong motivation to empower deaf individuals, enabling them to experience greater independence without relying on others, with an aim to bridge the gap between auditory awareness and the needs of the deaf and hard-of-hearing community. Keywords: Deaf and Hard of Hearing, Convolutional Neural Networks, Sound Classification, Spectrogram.
A Web application to detect phishing links using machine learning approach
(Strathmore University, 2023) Kimani, B. M.
Covid-19 fast tracked the transition of ways of working and learning into remote working models and with it the rise of Cyber Crime. Phishing is one of the most effective cybercrimes where attackers deceive users into falling for baits that enable them to steal personal information including identification or social security number, password, credit card details, debit card details and usernames which they then use to commit crimes mostly associated with financial losses and or identity thefts. These attacks are frequently carried out through malicious emails, texts, and phone calls. Even in the presence of robust antivirus software and phishing detection tools, phishing has continued to be rampant and widespread as adoption of IT across the world goes up. Several methods have been implemented to deal with phishing attacks including, blacklisting phishing Uniform Resource Locators (URLs), heuristic rule based and machine learning models. The blacklist anti-phishing approach is limited in its ability to detect new phishing URLs due to its reliance on a pre-compiled list of phishing URLs, whereas most Machine Learning (ML) methods for detecting phishing websites have been reported with very decent detection accuracy but significant false alarm rates. Attackers are continuously re-inventing methods of attacks which as well makes heuristic rule-based methods vulnerable to failed detection. This study has provided further information on phishing attacks to create awareness, designed and implemented a web-based phishing links detection tool using ML deep learning model which achieved an accuracy of 94.27 complemented by heuristic rules and blacklist capabilities. Further improvement areas including continuous models training and features re-engineering for improved prediction accuracy, and equipping Internet of Things (IoT) gadgets with developed Application Programming Interface (API’s) capability to determine what endpoints to or not to interact with. Keywords: Remote working, Phishing, Identity thefts, Machine Learning, Detection tool.
An Online stock market recommender system using machine learning
(Strathmore University, 2023) Muoki, S. M.
Investing in the stock market presented significant challenges for novice investors, primarily due to the overwhelming volume of available data. Consequently, novice investors often made irrational decisions and experienced unfavourable investment outcomes, particularly in the context of Kenya. This project aimed to address this issue by developing an innovative online stock market recommender system that utilised machine learning techniques. By leveraging these techniques, the system aimed to facilitate informed decision-making based on reliable data. Novice investors frequently encountered difficulties in accessing sufficient and well-organised information about the companies they intended to invest in, resulting in suboptimal returns. Traditional research methods often failed to adequately address the complexities and vastness of the stock market data. However, incorporating machine learning into the investment process held promise for analysing historical data, identifying patterns, and providing valuable insights to support informed decision-making. To comprehensively achieve the research objectives, a mixed-methods approach was employed, which integrated both quantitative and qualitative data collection and analysis in a sequential design. The Object-Oriented Analysis and Design (OOAD) technique was systematically and logically adopted to develop the software system. Additionally, the Dynamic System Development Methodology (DSDM) served as a guiding framework to address the identified problem and facilitate the development of the online stock market recommender system. This research project identified the information challenges faced by individual investors in the stock market, highlighting issues such as limited access to critical information, lack of necessary skills, and reliance on inaccurate media reports. Moreover, the study identified specific factors that significantly influenced stock market investments, including earnings per share, firm fundamentals, market trends, and news sentiment. To overcome these challenges, the project focused on developing an advanced online stock market recommender system that effectively leveraged machine learning techniques to provide personalised investment recommendations. The system underwent rigorous testing, demonstrating superior performance by offering accurate recommendations and comprehensive investment performance reports. Keywords: Investing; Stock market; Novice investors; Poor investment outcomes; Online stock market recommender system; Machine learning techniques; Informed decision-making; Information challenges; Mixed-methods approach; Quantitative and qualitative data; Object-Oriented Analysis and Design (OOAD); Dynamic System Development Methodology (DSDM); Earnings per share (EPS); Market trends; News sentiment; Investment performance reports
Cross-Lingual model for hate speech detection on Twitter: a case of Swahili and Swahili-English slang
(Strathmore University, 2023) Kariuki, A. O.
The prevalence and entrenchment of online hate, hate crimes and hate speech in contemporary society concern organisations and governments. Detecting online hate, especially on social media, has proven daunting as offensive languages have multifaceted behaviours, and most training data are topic specific. On top of that, available solutions and research are geared towards the English language; thus, detecting online hate in lower-level languages like Swahili and Indigenous African Languages is much more difficult. This has worsened because social platforms such as Twitter, Facebook, Instagram, Rumble and YouTube enable consumers to converse and participate in their native dialects. This research proposed using cross-lingual transfer learning for hate detection to overcome these challenges. A Cross-Lingual model built on a BERT pre-trained model was developed as part of the research's experimental methodology, and its performance was compared to those of more established text classifiers like SVM, NB, and LR. Through the Twitter API, more than 300K tweets with a Kenyan focus were collected. These tweets focused on Kenya's most divisive moments in history, namely the 2013, 2017, and 2022 general elections. A set of predetermined criteria, including user location, tweet location, hashtags, pro-hate accounts, hate patterns, and racial epithets, were used to collect the data. For usage in the model development, training and validation, a random sample of over 20K tweets was annotated as hate or non-hate. The developed Cross-Lingual model achieved a ROC curve area under the curve of 0.77 and an accuracy of 77 per cent. The following are the contributions made by this study. Primarily, the research established an empirical framework and methodology for utilising transfer learning to identify the offensive language in low-resource languages. Additionally, this strategy was crucial in creating a text classification framework that could be broadly applied to different types of abusive language on online platforms. The model's results may thus be used to inform data-driven legislation regarding the detection of online hate as well as evidence-based decisions by pertinent intelligence agencies. Keywords: Deep learning, free speech, freedom of speech, hate detection, hate speech, machine learning, natural language processing, social media, Twitter.
Customer churn prediction tool using deep learning: a case of an ecommerce business operating in Kenya
(Strathmore University, 2023) Narina, P.
The problem of customer churn poses significant challenges for businesses, particularly in the e-commerce sector. With a high level of competition and low customer retention rates, businesses face the risk of losing customers and experiencing a decline in revenue. Previous research in customer churn prediction exhibited limitations in terms of accuracy and the lack of consumer-facing tools for predictions. Moreover, existing studies were often conducted before the COVID-19 era, failing to capture the impact of recent changes in consumer behavior and market dynamics. This research aimed to develop an effective customer churn prediction model for e-commerce businesses in Kenya. The study aimed to address the limitations of existing research by leveraging advanced machine learning techniques and considering the specific challenges faced by businesses in the post-COVID-19 era. To achieve the objective, an agile software development methodology was adopted. This approach allowed for continuous iterations and refinements during the model development process. Customer dataset was obtained from Kaggle an online platform for sharing datasets. The dataset included customer demographic information, transaction history, and customer engagement metrics. The data was carefully pre-processed to handle missing values, outliers, and ensure data quality. The multilayer perceptron model (MLP), a powerful deep learning algorithm, was employed to train the customer churn prediction model. The dataset was split into training and testing sets, with an 80-20 ratio, to assess the model's performance. The results of the study indicated that the developed customer churn prediction model achieved high accuracy, with precision, recall, and F1 scores of 94%. This demonstrated the model's effectiveness in identifying potential churners and enabling businesses to take proactive measures for customer retention. The findings of this study have significant implications for e-commerce businesses, providing them with a valuable tool to predict customer churn and implement targeted retention strategies. By leveraging the power of advanced machine learning techniques, businesses could enhance customer satisfaction, optimize resource allocation, and drive sustainable growth in a highly competitive market. Keywords: Customer Churn, Deep Learning, Multi-Layer perceptron, E-commerce, Big Data.
Narrowband-Internet of Things based water metering system: a case-study for Nairobi County
(Strathmore University, 2023) Kamau, S. I.
The adoption of smart water metering systems is gaining interest in commercial and domestic sectors. This technology can monitor water usage and solve challenges like leakages and water theft. Water companies in Kenya have started looking for solutions to monitor water usage, especially in urban centers. The broad implementation of these meters is facing limitations such as poor infrastructure for network connectivity and short battery life, leading to downtimes in data transmission. This study designs and implements a smart water metering system that utilizes Narrowband-Internet of Things (NB-IoT) communication technology. NB-IoT communication technology is a classification of Low Power Wide Area Networks (LPWANs) connectivity designed for IoT devices. The study included integrative reviews to determine the current water meters and how current data is collected and processed. The implemented system ensured transparency in collecting water metering, continuity of water supply to homesteads, and reduced labor for the water vending companies in Kenya. The system has a combination of hardware and software that demonstrates how NB-IoT technology is suitable to facilitate solving the existing issues faced by Nairobi City Water and Sewerage Company (NCWSC). The collected metering data was sent to a central server in real-time. It was then processed and analyzed to determine usage patterns. In addition, a web interface was developed to assist in the visualization of information by clients. The system demonstrated exceptional reliability over the deployment period by providing water consumption data. The communication reliability was exponential as the NB-IoT connection was maintained even inside buildings in Strathmore University indicating penetrations through the walls hence easy to deploy in other areas beyond Nairobi. Keywords – Narrowband-Internet of things (NB-IoT), Internet of Things (IoT), Low Power Wide Area Networks (LPWAN)
Sexual and Gender-Based Violence reporting and pattern analysis tool: a case of Nairobi County
(Strathmore University, 2023) Makau, A. N.
Globally, sexual gender-based violence (SGBV) has affected individuals irrespective of age, gender, and social status. This harmful act includes sexual, mental, physical, and economic harm to both the private and economic sectors. SGBV has devastating consequences that can last a lifetime for survivors and even lead to death, hence the need for immediate medical attention. In the Kenyan context, SGBV continues to be a severe problem for society despite government efforts to address it through legislative and/or policy frameworks The inability of health, non-governmental organizations, and the government, as well as duty bearers, to combat SGBV has been impeded by a lack of up-to-date data. Governmental limitations in terms of planning to ensure that services are put in place to prevent SGBV and provide assistance to victims have attracted a rise in the number of affected victims. In this study, data was collected using questionnaires, and the collected data was used to find out the challenges in the reporting of SGBV cases and data management processes by the duty-bearers involved. The data was analysed using Microsoft Excel as a correlational tool, and the results were used to guide the researcher in the design of the mobile and web application. The study adopted an applied research design, and Agile methodology was used to develop the application. In order to understand the trends and patterns of SGBV, a KNCHR database with SGBV and feature-related information was first compiled and prepared to be used for ML modelling purposes, making use of the Poisson Regression Algorithm because it is best suited for identifying patterns and relationships between different factors in this case; gender, location, month, and year, and how these variables may affect the frequency of SGBV. Secondly, an android-based mobile application was developed to assist victims in reporting the violations in a timely manner, assist hospitals and police stations in data management by allowing them to update the details of the case with necessary information on the application, seek duty bearers’ attention and assist in the mapping of the reported cases for purposes of visualizing the number and geo-location of these cases, ensuring transparency throughout the ecosystem. The patterns depicted from the ML model were integrated with the web application for duty bearers to view the patterns of previously reported violations as well as a reports module on the web application displaying analysis in form of charts of the reported cases through the mobile app including the gender, age, type of violation and location. These analyses will aid in planning ahead for better resource optimization and comprehension of the magnitude of SGBV in order to mitigate future incidents.
Travel agency recommender system based on social media sentiment analysis
(Strathmore University, 2023) King'ori, S. W.
In today's highly competitive e-tourism industry, online reviews and recommendations are crucial to customers' travel decisions. This study focuses on reviewing the level of service quality in the e-tourism sector through social media sentiment analysis, aiming to aid travelers in making informed decisions about their travel arrangements through a recommender system. While other methods exist for recommender systems, they are not sufficient for the specific context of Kenya. To address this, the researcher develops a tool capable of providing personalized recommendations based on user destinations using the transformed data received from sentiment analysis. The study adopts an exploratory research design, targeting individuals who engage in social media discussions related to the e-tourism industry, including travelers, travel companies, and other tourists which is supported by Agile Development Methodology. The collected data is retrieved from Twitter, a prominent communication platform for travel agencies. The study utilizes SnScrape for tweet extraction and employs data preprocessing techniques to categorize and analyze the collected data using the Vader lexicon model. The collected data undergoes sentiment analysis, where each evaluated tweet is assigned a polarity tag indicating whether it is positive, negative, or neutral sentiment, with the analyzed results presented using charts and tables. Machine learning algorithms play a crucial role in providing personalized and relevant recommendations to users based on their destination preferences and historical data. K-Nearest neighbor, Support Vector Machine, and Naive Bayes are explored and evaluated to show the best-performing algorithm for the recommending system. A hybrid filtering approach is incorporated using content-based and collaborative filtering to create travel profiles based on the selected Reliability and validity measures are applied to ensure research quality, e research quality, both reliability and validity measures are applied which exhibit high levels of accuracy, precision, and F1 scores indicating their effectiveness in recommending travel agencies. Streamlit is used to build an interface and deploy the machine learning models using a set of rules to recommend travel agencies based on the user’s destination. The study concludes that recommender systems rely on feedback and online reviews shared by customers after traveling to various destinations. It recommends that travel providers acknowledge this trend and actively encourage customers to share their experiences through social media and other platforms. Additionally, the study suggests that users of sentiment analysis tools ensure diverse training data to mitigate bias and accurately reflect the sentiment of the target audience. Keywords: Service Quality, Sentiment Analysis, Travel Agencies, Recommender System, Twitter, Machine learning

Browse

Browsing MSIT Theses and Dissertations (2023) by Title

Results Per Page

Sort Options