Open Access

Binge Eating (BE) is a common disordered eating behavior, and the study explored social-cultural influences associated with BE development among university students in Nairobi, Kenya. Objectives included determining the influence of ethnicity, social comparison, mass media, peer and family pressure and the mediating role of coping orientation. A descriptive and correlational research design involving 395 participants from the University of Nairobi, responded to online close-ended survey questions from the Binge Eating Scale (BES), Social Comparison Scale (SCS), Coping Orientation to Problems Experienced Inventory (Brief COPE), Social Cultural Attitudes Towards Appearance Questionnaire-4 Revised (SATAQ-4R Male and Female) and a Socio-Demographic Questionnaire. Results showed that ethnicity and social comparison do not have a significant association with BE. Mass media has a significant association with BE among both males and females. The societal, peer and family pressure also have a significant association with BE among males and females. Coping orientation had a partial mediation on mass media, family and peer pressure among females. Similar partial mediation findings were present among males with regards to mass media and family pressure, but coping orientation had no mediating association with peer pressure. Findings reveal that social-cultural factors are associated with BE through direct and indirect pathways mediated by coping strategies, with notable gender differences in vulnerability patterns and coping mechanisms. Therefore, more advanced research in this area needs to be explored to understand BE and EDs from a Kenyan view point. Keywords: binge eating, peer pressure, family pressure, ethnicity, mass media, university students. 

The goal of this thesis is to use methods from automated reasoning to analyze graphs occurring in the context of wireless networks. For this a suitable representation of the considered graphs, the transformations and algorithms executed on the graphs and also the underlying geometric conditions are required. The considered graphs resp. classes of graphs in this thesis refer to a) graphs only defined by geometric conditions, b) graphs defined by more general axioms after applying transformations, c) graphs resulting from a graph transformation algorithm. For studying graphs in the context of wireless networks, an important task is the construction of graphs without intersecting edges. Such graphs are called plane graphs and have a wide range of applications, for example in routing procedures. For the construction of connected plane graphs on a given set of vertices, the concepts of Gabriel graphs and relative neighborhood graphs exist. These concepts were originally developed for the Euclidean plane with Euclidean distance. In this thesis it is investigated, for which more general distance functions, Gabriel graphs, and relative neighborhood graphs are plane graphs, whenever possible by using the hierarchical prover HPiLoT. In addition to graph constructions, this thesis also examines graph transformations. A transformation is simple if the existence of an edge uv in the transformed graph depends only on the existence of the edges uv and vu in the underlying original graph. In the original graph, predicates determine which edges must exist, which edges must not exist, and which edges exist depending on the existence of other edges. It is investigated under which conditions for the predicates and for which transformations expressions can be found to automatically examine the containedness relation between two graph classes, whereby only graphs up to a certain number of vertices need to be considered. For the expressions found in this way, the containedness relations are evaluated for all possible graph classes using Mathematica. Graphs can also be transformed using algorithms instead of simple transformations. This thesis examines a planarization algorithm that generates a plane graph from an original graph by selecting a subset of edges. To maintain, in addition also the graph’s connectivity, the original graph must satisfy certain structural properties. Such properties are the redundancy and coexistence property. It is shown that for RCGs, i.e. graphs that satisfy these two properties, a connected plane subgraph always exists. To obtain this connected plane graph, the correctness of a local planarization algorithm for RCGs is proven by verifying the individual proof steps of a manual proof using the theorem prover Z3. To perform this verification, a suitable axiom system is defined that describes, the properties of the graphs, the properties required for the proof, as well as the underlying geometry.

Clustering algorithms partition unlabelled data into groups based on similarity, yet standard methods provide no guarantee that different demographic subpopulations are treated equitably across those groups. A second, often overlooked gap is interpretability: the partition produced by an algorithm such as k-means gives no human-readable account of why a data point was assigned to its cluster. This thesis addresses both gaps jointly by building a unified evaluation framework in which fairness and explainability are assessed alongside conventional clustering quality. The framework implements and compares four clustering methods k-means, k-medians, a fairlet-based fair k-medians method derived from quadtree decomposition, and a bounded-representation method across four real-world datasets: Bank Marketing, Adult Census Income, COMPAS Recidivism, and German Credit. Each configuration is evaluated on three axes: clustering quality (silhouette score, sum of squared errors, PAM cost), group fairness (Balance, violation rate, and cost of fairness), and explainability (surrogate decision-tree fidelity and complexity). Protected attributes reflect established fairness benchmarks: marital status for Bank Marketing, gender for Adult, race for COMPAS, and gender for German Credit. The results show that enforcing fairness does not uniformly degrade clustering quality. The bounded-representation method achieves a cost-of-fairness ratio close to 1.0 across all four datasets, meaning the quality penalty for a formal fairness guarantee is small in practice. Surrogate decision trees maintain high fidelity on both constrained and unconstrained clusterings, showing that post-hoc explainability is compatible with fairness-aware assignment. A recurring finding is that k-medians produces more balanced clusters than k-means without any explicit fairness constraint — a structural effect of the L1 objective yet it offers no provable guarantee. This contrast illustrates why formal methods such as bounded representation remain necessary. Together, the results provide an empirical basis for choosing clustering methods when both accountability and interpretability are required.

Recent progress in intelligent, adaptive, and security-aware Software-Defined Wireless Sensor Networks (SDWSNs) is driven by the integration of Machine Learning (ML) with Software-Defined Networking (SDN) and Wireless Sensor Networks (WSNs). This systematic survey analyzes 46 peer-reviewed articles published between 2024 and 2025, providing a problem-oriented synthesis of ML-SDWSN research. Emphasizing security, routing, and performance optimization, with a particular focus on deployment architectures, the survey identifies three major trends: increased adoption of ensemble and Reinforcement Learning (RL) methods for security and adaptive control; broader implementation of edge-based ML to minimize inference latency; and greater emphasis on privacy-preserving techniques, especially Federated Learning (FL). The survey presents a structured taxonomy encompassing seven thematic areas: Distributed Denial-of-Service (DDoS) mitigation, Intrusion Detection Systems (IDSs), routing optimization, Quality of Service (QoS) management, privacy preservation, data integrity, and network-efficiency optimization. Findings are synthesized from over 120 experimental configurations reported in the literature. Due to substantial differences among the reviewed studies in terms of datasets, network topologies, hardware platforms, measurement definitions, and validation methodologies, the reported values are presented as descriptive cross-study aggregates rather than direct comparative benchmarks or formal effect-size estimates. Within these constraints, the survey identifies recurring trade-offs among accuracy, latency, scalability, and privacy. It provides evidence-based design considerations for researchers and practitioners. The survey also highlights eight critical research gaps, including limited multi-dataset validation, a lack of real-world deployments, insufficient scalability analysis, and the need for rigorous evaluation of RL-based SDWSN control.