SU+ Digital Repository

SU+ is an online repository for the preservation and promotion of assorted digital content at Strathmore University

Off-Campus Access to restriced resources (including the ExamsBank) now requires registration using an @strathmore.edu email address

Authentication is NOT required for On-Campus Access to content

[ISSN 2519-5883]
Photo by @Strathmore University
 

Communities in DSpace

Select a community to browse its collections.

Now showing 1 - 5 of 7

Recent Submissions

No Thumbnail Available
Item
A Bi-Lingual counselling chatbot application for support of Gender Based Violence victims in Kenya
(Strathmore University, 2024) Mutinda, S. W.
Gender-based violence (GBV) remains one of the highest prevailing human rights violations globally, surpassing national, social, and economic boundaries. However, due to its nature, it is masked within a culture of silence and causes detrimental effects on the dignity, health, autonomy, and security of its victims. The prevalence of GBV is fuelled by cultural nuances and beliefs that justify and promote its acceptability. The stigma surrounding GBV in addition to fear of the consequences of disclosure deter victims from seeking help. Additionally, the resources available for addressing GBV such as legal frameworks and recovery centres are limited. Technological approaches have been established to tackle GBV as intermediate and supplementary support for victims as part of UN-SDG 5. Conversational Agents such as Chomi, ChatPal, and Namubot have been developed for counselling of GBV victims who struggle with disclosing their predicament to humans. The existing chatbots, however, are not a fit for Kenyan victims because they utilize languages such as Swedish, Finnish, Isizulu, Setswana and Isixhosa in addition to incorporating referral services specific to their regions. This research addressed this gap by developing a chatbot application suitable for the Kenyan region for counselling of GBV victims using both Kiswahili and English, the languages predominantly used in the country, in addition to including contacts to referral services within the country. The methodology utilized involved the development of a chatbot application based on Rasa open source AI framework by training a model using a pre-processed counselling dataset. The performance of the model was evaluated using NLU confidence score to determine the model’s certainty in its intent identification and a confusion matrix was generated which with 80% and 20% training and testing data split resulted in 100% classification threshold accuracy. Python’s Fuzzy Matching Token Set Ratio score was also used to determine the response which best matches the input with results indicating satisfactory performance of the model ranging between 63% and 92% for GBV queries input. The developed model was then integrated into a web application as the user interface for user access and interaction with the model hence achieving the research objective of developing a chatbot application to conduct counselling for GBV victims in Kenya using English and Kiswahili languages . Keywords: Gender-based Violence, stigma, chatbot, Rasa open source, NLU Confidence Score, Fuzzy Matching Token Set Ratio score
No Thumbnail Available
Item
Application of fingerprint authentication to fortify child safety in school transport
(Strathmore University, 2024) Mutuku, S. W
Safety of school-going children has been a great concern to parents, school administrations and the transport team in the recent past. In urban areas like Nairobi where most parents are busy working and crime is fast increasing, the need for an efficient and safe transport for pupils cannot be underestimated. Most current school transport systems use NFC cards or manual attendance records to keep track of the children picked in the morning or dropped after school. Using manual attendance is time consuming, especially where there are many students. NFC cards could also be lost or misplaced. This could be a security loophole if picked by someone else and manage to access the transport. This research uses fingerprint authentication for both learners and staff where fingerprints are captured, and database queried to authenticate the learner or staff. The choice of technology is inspired by the fact that fingerprints are unique to every individual adult or child. The research used Rapid Application Development (RAD) methodology because it is more flexible in accommodating the changing nature of requirements which are not well defined in the initial stages. The requirements are implemented in the system in separate prototypes until the final prototype is developed. It also allows for fast user feedback and speeds up delivery. Learners’ existing records will be used as input to the system and will be incorporated with the children fingerprint then stored in a database. Convenience sampling was used in the research to obtain simulated data. Keywords: Biometrics, safety, fortification, school transport, Facial Emotion Recognition, Biometric Fingerprint scanner, Geofencing
No Thumbnail Available
Item
A Customer churn prediction and corrective action suggestion model for the telecommunications industry using predictive analytics
(Strathmore University, 2024) Wanda, R. K.
The telecommunications industry is significantly susceptible to customer churn. Customer churn leads to loss of customer base which leads to reduction in revenue, reduced profit margins, increased customer acquisition costs and loss of brand value. Mitigating the effects of customer churn has proved to be a tall order for many organizations in the telecommunications industry. Most companies employ a reactive approach to customer churn and thus do not take any corrective actions until the customer has left. This approach does not enable organizations to know and prevent potential churn before it occurs. Alternatively, some organizations employ a more proactive approach to mitigate customer churn through predictive analytics. Although this approach is more effective, it only predicts which customers will churn without recommending the appropriate corrective action. In this dissertation, a customer churn prediction and corrective action suggestion model using predictive analytics was implemented to predict churn and suggest appropriate corrective actions. The IBM telco customer churn dataset accessed via API from the open machine learning.org website was used for this study. The dataset was subjected to pre-processing and exploratory data analysis to gain valuable insights into the data. To enhance the reliability of the developed model, an 80/20 train/test split was applied to the dataset. The training dataset was then divided into 5 folds before model fitting. Several classification algorithms; Logistic Regression, Gaussian Naive Bayes, Complement Naïve Bayes, K-NN, Random Forest and CatBoost were then fit with the training data and their performance was evaluated. Logistic Regression achieved a recall of 80% and was selected for system implementation. Logistic regression feature coefficients were then used to determine the appropriate corrective actions. A locally hosted web interface was then developed using the Python Streamlit library to enable users to feed input into the model and get churn predictions and corrective action suggestions. The developed model demonstrated ease of use and high performance and will enable telecommunication companies to accurately predict customer attrition and take appropriate corrective actions, reducing customer attrition's impact on the companies’ bottom line. Keywords: churn, machine learning, predictive analytics, telecommunications industry
No Thumbnail Available
Item
A Credit scoring model for mobile lending
(Strathmore University, 2024) Oindi, B.
An exponential increase in mobile usage has led to more accessible access to mobile loans for most Kenyans; this has created a lifeline for those excluded by traditional financial institutions; the easier way to borrow loans comes with its risks. The major one is borrower defaulting. This creates a need for credit scoring, which plays a crucial role in decision-making for lenders to determine borrowers’ creditworthiness, therefore minimizing credit risk and managing information asymmetry. On mobile lending, borrowers’ financial information is usually limited, making machine learning a favorable tool for credit assessment. Traditionally, the process required statistical algorithms and human assessment, which fall short when subjected to large datasets and are time-consuming. The traditional methods also need help adjusting to changes in borrowers' behavioral needs. Against this backdrop, this research developed a novel credit scoring model for mobile lending using Random Forest, XGBoost, LightGBM, Catboost, and AdaBoost algorithms. SMOTE was used to address the class imbalance problem. The model achieved the best accuracy of 86%. The research further analyzes the challenges in credit scoring and reviews related works by several authors. The research also looked at the feature importance of the models, which effectively analyzed the model's behavior. This model can analyze vast volumes of data, which would otherwise be resource-intensive if done manually. The machine learning model was then deployed into a Streamlit Web Application with a user interface where real-time predictions are made based on borrower data. The model can give lenders insights into determining borrowers' creditworthiness and enable them to make informed decisions before lending. Keywords: Mobile loans. Credit Scoring. Probability of Default. Machine Learning. Statistical Algorithms. SMOTE
Thumbnail Image
Item
Drowsiness detection system during driving using IoT and Machine Learning
(Strathmore University, 2024) Somo, A. M.
The interest in implementing drowsiness detection systems through the integration of IoT and Machine Learning, especially in the automotive and transportation sector is growing significantly. By utilizing this technology, it becomes possible to monitor and identify instances of driver drowsiness, addressing safety concerns related to fatigue related accidents. However, the widespread adoption and application of these drowsiness detection systems encounters some challenges such as poor telecommunication for network connectivity for IoT devices and ensuring efficient resource utilization within the constraints of Machine Learning. These are the main challenges faced by drowsiness detection systems during driving. This study designs and implements an efficient drowsiness detection system that utilizes Machine Learning and IoT technologies. The approach will involve the deployment of an IoT connected sensor, which is a camera within the vehicle’s environment. This sensor will collect real-time data on the driver’s eye movements. This raw data is then preprocessed to extract the relevant features and then processed information will be fed into the Machine Learning model. This model, which is optimized for low-resource environments will be able to perform real time drowsiness classification. Our model will employ CV2, KNN and Dlib algorithms independently. The purpose of implementing these distinct machine learning algorithms is to conduct a comprehensive assessment and comparison of their performance. By doing so, we will be able to determine which algorithm yields the best results in terms of accuracy, thus enabling us to make an informed decision. The implemented solution will aim to enhance transparency and consistency in the acquisition of drowsiness related data. This initiative will make things easier for drivers and demonstrate how we can use IoT and Machine Learning technologies to solve problems around detecting drowsiness. By using both hardware and software, the system will show how we can use IoT concepts to solve common problems in drowsiness detection. The hardware we're using includes a computer camera as the sensors, and we'll also use the OpenCV framework libraries to train the machine learning model. The collected data associated with the drowsiness levels will then be transmitted to a central server for real time analysis. The data will undergo thorough processing and assessment to identify patterns of drowsiness instances. Furthermore, a User-friendly python interface will be developed to provide clients with visual insights into the detected drowsiness instances. Keywords – Internet of things (IoT)