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

Photo by @inspiredimages

Communities in DSpace

Select a community to browse its collections.

Now showing 1 - 5 of 7

Recent Submissions

  • Thumbnail Image
    Item type:Item,
    Sentiment analysis on Swahili - English code switched tweets via transfer learning
    (Strathmore University, 2024) Kibiru, G. J.
    Sentiment analysis is a technique that is used to determine the sentiment, or emotional content, of a piece of text. When applied to code switched data, sentiment analysis can be used to determine the sentiment of text that contains words from multiple languages. This is a challenging task, as code switching can introduce complexity and ambiguity into the text. This study will present the use of transfer learning for sentiment analysis on Swahili-English codeswitched data using deep neural network models. This study will focus on the use of transfer learning in conducting sentiment analysis on Swahili-English code switched data. The study will consider two pre-trained deep learning algorithms, that is mBERT and SwahBERT. This study will use these pre-trained deep learning models to conduct sentiment analysis on Swahili-English code switched tweets gathered between the period 29th March 2022 to 15th August 2022 and compare their performance using accuracy, specificity, precision, recall and f1 score metrics.
  • Thumbnail Image
    Item type:Item,
    Leveraging clustering for improved marketing strategy in e-commerce: a customer lifetime value approach
    (Strathmore University, 2024) Gichuyia, K. K.
    In today's dynamic business landscape, characterized by a shift towards service-focused economies, companies are experiencing a transformative paradigm. They are proactively adapting to a new era, emphasizing the cultivation of enduring customer relationships as the linchpin of sustainable profitability. This strategic shift underscores the pivotal role of marketing, which extends beyond traditional paradigms to serve as the cornerstone for enhancing a company's financial performance. Within this context, marketing endeavors are geared towards augmenting what we refer to as "Customer Lifetime Value" (CLV), a multifaceted concept akin to a mosaic, encapsulating the cumulative value derived from loyal customers over time. Various models, including the commonly used RFM (Recency, Frequency, Monetary) model, have been utilized in predicting customer lifetime value (CLV). The RFM model evaluates customers based on the recency, frequency, and monetary value of their transactions. Additionally, conventional methods like the widely used Elbow approach have been employed to determine the optimal number of clusters in CLV models. However, this study aims to explore CLV, particularly within the E-Commerce sector, by leveraging the analytical power of the Single Value Decomposition (SVD) clustering method. The paper underscores the critical significance of CLV models in navigating this intricate domain. These models serve as potent instruments for segmenting the market intelligently and optimizing resource allocation for marketing activities. In E-Commerce, where strategic decision-making is vital, businesses deploy these resources judiciously to acquire, retain, and cross-sell to customers, epitomizing the astute acumen required for E-Commerce success. In the realm of E-Commerce, it has been customary to assess Customer Lifetime Value through the prism of Recency, Frequency, and Monetary (RFM) variables. However, it is essential to recognize that the relative importance of these variables undergoes dynamic interactions influenced by product or service attributes and industry-specific idiosyncrasies within the E-Commerce domain. To encapsulate, this paper delves into the intricate facets of CLV, unveiling the potential of various CLV models empowered by the clustering method, using Single Value Decomposition approach to determine the most optimal clusters, as strategic assets for modern E-Commerce. These models serve as a compass for market segmentation and resource allocation, thereby sculpting the trajectory towards success for E-Commerce enterprises in the ever-evolving landscape of customer-centric commerce. Key words: Customer Lifetime Value, Recency, Frequency and Monetary Model, Customer Relationship Management, Prediction
  • Thumbnail Image
    Item type:Item,
    Driver drowsiness detection in the freight industry
    (Strathmore University, 2025) Okero, M.
    Driver drowsiness has over the years become a key concern to everyone involved in long distance travels especially in the freight industry. Year in, year out, the number of deaths and fatalities globally as a result of driver drowsiness keep increasing significantly. Thus, ensuring the road safety of people is of uttermost importance. One of the safety measures employed against driver drowsiness is the use of a dashboard camera. Despite the widespread adoption and growing numbers of installations of dashboard cameras (Dashcams) across the globe with even evolved technology, current dashcams are still incapable of learning how to identify different postures or gestures that indicate that a driver seems to be either distracted, drowsy or asleep while driving. Even though they have the capability of recording anything happening on the road or in the vehicle in the event of an accident, they are still unable to provide real-time warnings or triggers to the drowsy driver attempting to possibly prevent an accident from happening. They also require continuous monitoring which is ineffective due to a human’s inability to maintain sufficient attention to discern significant events, a gap that the proposed enhancement aims to fill. The proposed Machine Learning Aided Drowsiness Detection System intends to cater for the fundamental flaws in today’ dashcams. Machine learning incorporates aspects of artificial intelligence that empower systems with the ability to continuously learn and improve automatically with experience without being explicitly programmed. It triggers a new way of thinking about the current dashcams. It offers new features, such as real-time conscious monitoring and gives an alert to the driver in the case of drowsiness being detected, in addition to the pre-existing systems’ features – a visual system that not only ‘sees’, but also ‘understands’ what it’s ‘seeing’. It is an undeniable fact that the use of dash cams integrated with machine learning will offer new robust capabilities. The Drowsiness Detection System comprises co-working components of a computer vision, camera, and a special type of machine learning model based on Deep Learning using Neural Networks which excels in object detection and recognition tasks via image analysis achieving an accuracy of 96.19%. The drowsiness detection system is thus highly efficient in the identification of drowsiness from different facial features such as eyes and mouth, send out a warning or alert in the event that drowsiness is detected and be continuously trained and improved with better and more datasets. The proposed system is convenient due to its improved performance and efficiency.
  • Thumbnail Image
    Item type:Item,
    Spatial modeling of the association between distance to hospital emergency care and severe anemia among children aged 1−59 months in Busia County
    (Strathmore University, 2024) Mutinda, M. M.
    Background: Access to Emergency Care (EC) services is a core component towards ending preventable paediatric deaths under the SDG 3.2. Physical access to EC is one of the facets of access that has been shown to be associated with health outcomes. Standard regression models often used to assess the association have key limitations including failure to adjust for either spatial heterogeneity in the risk of outcomes or spatial autocorrelation in outcome incidence. This study aimed to develop a Bayesian Model-Based geostatistical model to assess the association between physical access to EC services and severe anaemia among anaemic paediatric admissions in Busia County Hospital using the INLA-SPDE framework and compare changes in the observed association with results from the standard logistic regression. Methods: Data from a hospital surveillance for paediatric admissions aged 1−59 months who reside in a malaria endemic setting and were anaemic were assembled. Four models were fitted, two under the INLA-SPDE framework and two under the standard logistic regression framework. Physical access was defined as village travel times to the county hospital and adjustments for known confounders were done including spatial variations in Plasmodium falciparum Prevalence Rate (Pf PR) as the underlying driver of anaemia. Differences in the travel time coefficients were assessed across the models. Results: In the developed model, INLA-SPDE model with spatially varying coefficient for Pf PR, the association between physical access to EC services and severe anaemia was significant only among admissions within 30−59 minutes of travel time (AOR: 1.94, 95% CI:1.18−3.08) when compared to admissions within < 30 minutes of travel time to Busia County Hospital. In the standard logistic regression models and standard INLA-SPDE model, the risk of severe anaemia was associated with poor physical access across all other admissions in comparison to admissions within < 30 minutes of travel time. However, coefficient confidence intervals under the standard INLA-SPDE model were wider compared to those in the standard logistic regression models. Conclusions: In assessing the association between physical access to EC services and health outcomes, it is vital to not only adjust for spatial heterogeneity in the underlying drivers of health outcomes, but also to appropriately model the association. Further, in the presence of spatial dependence, models should account for spatial autocorrelation so as not to underestimate standard errors. KEY WORDS: Spatial heterogeneity, Spatial autocorrelation, INLA-SPDE, Anaemia, Malaria, Travel time
  • Thumbnail Image
    Item type:Item,
    Delineation of residential housing submarkets using spatially constrained multivariate clustering
    (Strathmore University, 2024) Njoroge, S. N.
    Every housing market is made up of unique submarkets. Submarkets are areas or neighborhoods where houses have similar features, such as the age of the houses or price. Segmenting housing markets into submarkets is recommended for better understanding and more effective interventions in the housing market. While different submarket delineation approaches exist, many do not impose spatial constraints, overlooking the spatial relationships between houses. This oversight results in submarkets with poorly defined boundaries that do not match the urban layout, making accurate spatial inferences difficult and limiting stakeholders' ability to establish policy zones. To address these limitations, this study uses the SKATER clustering algorithm, which demarcates submarkets by taking into account the location of houses and ensuing spatial relationships alongside the structural attributes of the houses. The proposed method is implemented in a case study of King County, using house sale data from May 2014 to May 2015. It identifies four submarkets with boundaries closely aligned with the landscape, marking improvement over previous research. The analysis reveals a notable housing market imbalance whereby northern cities like Bellevue feature high-priced, spacious, high-quality houses on large lots. At the same time, the southern region, including SeaTac and Federal Way, offers older, smaller houses at relatively lower prices. These findings help stakeholders and investors make accurate spatial inferences for addressing housing challenges, particularly market imbalances. Keywords: housing market, spatial constraints, submarkets