The Nelson Mandela African Institution of Science and Technology
Institutional Repository
Research Collection
Select to browse Research collections.
Recent Submissions
Unlock the drivers of early ANC visits among pregnant women in Kasulu town council, Tanzania: an institutional cross-sectional study
(Springer Nature, 2025-10-07) Lyoba, Winfrida; Mpambije, Chakupewa; Mwakatoga, Joyce
Background Maternal and child mortality remains a global public health challenge. Thus, countries, including
Tanzania, have adopted different cost-effective models, especially antenatal care (ANC) to improve maternal and child
health (MCH). Despite the early timing of ANC visits having a great implication for ensuring improved MCH services,
Tanzania has disproportionately experienced late ANC visits among pregnant women. This has entailed conducting
an institutional-based study in Western Tanzania, Kasulu Town Council (KTC) to ascertain whether demographic socio
economic and maternal characteristics imply the persistence of late ANC visits using robust methods.
Methods An institutional cross-sectional study design was conducted in KTC, Kigoma Region using an embedded
mixed-method approach from March-April 2020. Quantitative data was collected from 320 women with children
aged 0–6 months attending postnatal services. A total of 40 participants were involved in the qualitative study
through in-depth interviews with healthcare providers and focus group discussions held with pregnant women and
women with children aged 0–6 months. Descriptive statistics and multivariate binary logistic regression were used
to determine the characteristics associated with the timing of ANC visits among pregnant women. Furthermore,
thematic analysis was used to generate themes triangulated with quantitative results.
Results Findings revealed that 32.2% of pregnant women attended ANC visits in the first trimester. Early ANC was
associated with maternal age (AOR = 1.839, 95% Cl: 1.023, 3.303), being accompanied by a partner (AOR = 2.165, 95%
Cl: 1.256, 3.733), and awareness of the danger signs (AOR = 2.079, 95% Cl: 1.172, 3.687) and parity (AOR = 2.164, 95%
Cl: 1.091, 4.291). Little association was noted in the knowledge of ANC timing (AOR = 0.564, 95% Cl: 0.320, 994) and
household income (AOR = 0.529, 95% Cl: 0.281, 0.995). Qualitative data indicated that low rate of early ANC initiation
was attributed to a lack of support from partners and accompanied to ANC visits, insufficient knowledge of the timing
of early ANC visits, and socio-cultural beliefs.
Conclusion Results confirmed that early ANC visit in KTC is low. Revealed associated factors act as a bridge to
improve maternal and newborn health and contribute to achieving Sustainable Development Goal no 3, which
targets maternal mortality of less than 70 deaths per 100,000 live births and neonatal mortality of 12 per 1000 live
births by 2030. Proposed integrated interventions can potentially ensure that women, regardless of pregnancy status,
A Hybrid Deep Learning Model for SQL Injection Attack Detection
(IEEE Access, 2026-01-12) Casmiry, Emanuel; Sinde, Ramadhani; Mduma, Neema
An increasing number of web application services raises significant security concerns. Online access to these applications exposes them to multiple cyberattacks. The Open Web Application Security Project has consistently reported SQL injection attacks among the top 10 cyber threats for over a decade, highlighting the need for more effective detection and prevention strategies. Traditional detection methods, primarily signature-based, offer basic protection but struggle to identify new signatures embedded within web requests. An alternative involves Machine Learning (ML) and Deep Learning predictive analytics, which provide effective solutions for analyzing large datasets to detect and prevent SQL Injection Attacks (SQLIA). However, these models often face challenges such as high false alarm rates and low accuracy. This study proposes a hybrid Convolutional Neural Network–Long Short-Term Memory (CNN-LSTM) model for SQL injection detection. Five architectures were compared: Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), hybrid CNN-LSTM, and hybrid CNN-GRU. Among them, CNN-LSTM achieved the best results, with 99.85% accuracy, 99.86% precision, 99.85% recall, and a 99.85% F1-score, while also recording the lowest false positive rate (0.06%) and misclassification rate (0.15%). CNN-GRU, CNN, and GRU followed closely, though GRU reported the highest false positive rate (0.16%). The standalone LSTM model performed slightly weaker, with 99.80% accuracy, 99.80% precision, 99.79% recall, and 99.80% F1-score, along with false positive and misclassification rates of 0.06% and 0.20%, respectively. Overall, the findings show that all models achieved near-perfect performance, with hybrid architectures outperforming single-architecture models.
Dietary Niche Partitioning in Sympatric Afro-Alpine Rodents: Habitat and Seasonal Dynamics in Lophuromys flavopunctatus and Stenocephalemys griseicauda
(John Wiley & Sons, Inc, 2025-12-11) Abraham, Birara; Katakweba, Abdul; Stella, Kessy; Ngoda, Upendo; Christopher, Sabuni; Rija, Alfan
Dietary niche partitioning is an important strategy for species to reduce competition, but information on how two sympatric rodents, Lophuromys flavopunctatus and Stenocephalemys griseicauda, within the Guassa Menz Community Conservation area co-exist is not known. Dietary composition, seasonality and habitat-specific resource use within Festuca grassland, shrubland and swamp grass habitats were determined by examining stomach content through micro histology. The findings were that both species utilised diets with a main component of vegetative plants; L. flavopunctatus, however, the diet had a much greater invertebrate component, while S. griseicauda used plant material preferentially, especially in Festuca grassland environments. Seasonal variation in diet was highly conspicuous, and wet season diets were made up of higher proportion of invertebrates and fruits compared to the dry season diets, which were predominantly vegetation. Further, habitat in swamp grass exhibited greater dietary variation, which is an indication of resource availability. In contrast, niche overlap was, nonetheless, consistently high (0.86–0.98), reflecting low interspecific competition, perhaps mediated by microhabitat or temporal resource partitioning. These results highlight habitat heterogeneity and seasonally fluctuating resources as important determinants of coexistence in Afro-alpine communities. The research highlights the ecological vulnerability of these species to environmental perturbations and adds to a better understanding of niche dynamics in Afroalpine rodent assemblages.
An integrated Conceptual Framework to Harness Artificial Intelligence for Enhanced Data Management in Context
(Journal of Information Systems Engineering and Management, 2026-01-31) Mosha, Neema; Ngulube, Patrick
This study explores the transformative impact of artificial intelligence
(AI) technologies on research data management (RDM) services in
higher learning institutions (HLIs), with a specific focus on the roles
played by academic libraries. It highlights how these libraries
strategically incorporate AI to enhance RDM effectiveness and
stewardship. An integrated conceptual framework is proposed, which
includes AI tools, RDM services, the functions of academic libraries,
necessary support structures, and the challenges related to AI and RDM
in HLIs. The findings indicate that this framework can significantly
boost the efficiency of AI adoption, leading to streamlined workflows
and reduced time for RDM services such as data generation, processing,
storage, access, and reuse. Additionally, the framework supports the
development of innovative AI technologies tailored to library users,
including research management software, data security and quality
assurance, and metadata generation.
Targeting Human Thymidylate Synthase for Chemotherapy: Application of a Computational Approach for Drug Repositioning and Natural Products Drug Discovery
(NM-AIST, 2025-08) Mteremko, Denis
Many solid tumors are treated with 5-fluorouracil and its analogs. However, 5-fluorouracil chemotherapy has been linked to resistance and significant toxicity. Using a computational approach, natural products were researched in silico in this study for their potential to be developed into anticancer drugs as an alternative to toxic chemotherapy. In addition, the US Food and Drug Administration (US FDA) drugs with potential for repurposing into useful drugs for treatment of solid tumors were also identified using computational drug discovery approaches. The US FDA drugs that have the potential to be further developed for treatment of numerous cancers for which 5-fluorouracil and analogs have been used as chemotherapy were identified through their repurposing against human thymidylate synthase (hTS). Flexible
receptor and ligand approaches were implemented. Virtual screening and hydrated docking approaches were applied to 2115 FDA drugs, with seven prioritized candidates that showed potential for repositioning. These included Erismodegib, Irinotecan, Conivaptan, Ergotamine, Lumacaftor, Imatinib, and Naldemedine. Furthermore, analysis of results also demonstrated that water plays a part in mediating protein-drug interactions. To obtain more precise estimations of the binding affinity more robust MM-PBSA and LIE approaches were applied. The MM-PBSA calculations for the FDA drugs Imatinib, Lumacaftor, Naldemedine, Erismodegib, Irinotecan and Ergotamine, revealed the free binding energy scores of − 130.7 ± 28.1, − 210.6 ± 29.9, − 238.0 ± 25.4, − 104.06 ± 19.97, − 131.88 ± 22.50, and − 66.7 ± 1.81 kJ/mol, respectively. These binding free energy scores revealed that Irinotecan, Lumacaftor and Naldemedine were the best binders. The analysis of molecular dynamics trajectories of the selected candidates revealed that, their Holo structures were overall stable. In addition, the consensus binding affinity scores from AutoDock Vina, LeDock and hydrated docking implementation of AutoDock 4.2 extension accounting for displaceable waters, and MM-PBSA approaches applied on both ensemble-based and crystal structures led to the choice of 1,2-dihydro-3-oxo-costic acid guaianyl ester 3 beta-O- (1,2-didehydro-3-oxo-costoyloxy)-
4beta,10beta-dihydroxy-guaia-1(2)-en-6beta,12-olide and 3beta-O-(1,2-didehydro-3-oxo- costoyloxy)- 4beta,10beta-dihydroxy-guaia-1(2)-en-6alpha,12-olide with the highest binding affinity in the picomolar order at 105.17 and 623.56 pM respectively, and good ADMET features.