<?xml version="1.0" encoding="UTF-8"?>
<rss xmlns:dc="http://purl.org/dc/elements/1.1/" version="2.0">
  <channel>
    <title>OAR@UM Community:</title>
    <link>https://www.um.edu.mt/library/oar/handle/123456789/14102</link>
    <description />
    <pubDate>Mon, 16 Nov 2026 16:27:07 GMT</pubDate>
    <dc:date>2026-11-16T16:27:07Z</dc:date>
    <item>
      <title>Novel data analytics techniques for analysing satellite imagery data, traffic and environment pollution data to predict the hospital admission rates</title>
      <link>https://www.um.edu.mt/library/oar/handle/123456789/150522</link>
      <description>Title: Novel data analytics techniques for analysing satellite imagery data, traffic and environment pollution data to predict the hospital admission rates
Abstract: Accurately forecasting hospital admissions is critical for optimizing healthcare resource allocation and emergency preparedness. This study presents a comprehensive, data-driven framework for predicting emergency department (ED) admissions related to respiratory, circulatory, and all-cause health conditions using local ground stations and satellite-based meteorological and air pollution data. Despite growing evidence linking environmental factors to hospital admissions, traditional hospital management systems lack integration with environmental intelligence and encounters problem in utilizing resources. To bridge this gap, our research employs advanced machine learning and deep learning models including tree-based, neural network, and regularized regression techniques to capture complex, nonlinear relationships between environmental exposure and health outcomes. A key contribution of this work is the introduction of three novel prediction schemes: Deep Explainable AI For Respiratory-Based Prediction (DXAIRP) for respiratory admissions, Deep Explainable AI For Circulatory-Based Prediction (DXAICP) for circulatory admissions, and Explainable Extreme Gradient Boosting Scheme for Environmental Impact on All-Cause Emergency Admissions (BGXAI) for all-cause admissions. These schemes were rigorously evaluated across multiple datasets, demonstrating high predictive performance and reliability. The analysis revealed that air pollution parameters (PM2.5, PM10, CO, NO2, and O3) are the most significant predictors of hospital admissions, while meteorological factors such as wind speed, temperature, and relative humidity also contribute, albeit to a lesser extent. The study further identified optimal lag times for different admission types, with a two-day lag providing the highest accuracy for respiratory and circulatory admissions, and a three-day lag optimizing all-cause admission predictions. Additionally, results indicate that ground-based environmental data yield superior predictive accuracy compared to satellite-derived measurements. To ensure interpretability and trust in AI-driven predictions, the framework integrates SHapley Additive exPlanations (SHAP), an Explainable AI (XAI) technique that elucidates the influence of each environmental factor on model outputs. The Extreme Gradient Boosting Regressor (XGBR) and Multi-Layer Perceptron Regressor (MLPR), and their ensembling with XAI establishes a robust, transparent, and adaptable predictive approach. Beyond its academic contributions, this research has significant practical implications. The predictive framework can be integrated into hospital management systems to support data-driven decision-making, enabling healthcare administrators to anticipate ED surges and optimize resource distribution. This framework includes ground station-based data to capture localized environmental conditions and also integrates satellite-based environmental data, making it particularly valuable for regions with limited ground monitoring infrastructure, thereby addressing spatial and temporal gaps in environmental health monitoring. Overall, this research advances climate-health analytics by offering a scalable, interpretable, and data-driven methodology for predicting hospital admissions based on environmental factors. The findings contribute to the growing field of AI-driven healthcare forecasting, with potential applications in national health systems, expansion to other disease categories, and deployment in resource-limited settings through cloud-based infrastructures.
Description: Ph.D.(Melit.)</description>
      <pubDate>Wed, 01 Jan 2025 00:00:00 GMT</pubDate>
      <guid isPermaLink="false">https://www.um.edu.mt/library/oar/handle/123456789/150522</guid>
      <dc:date>2025-01-01T00:00:00Z</dc:date>
    </item>
    <item>
      <title>Feedback loops and bias in machine learning algorithms for predictive policing</title>
      <link>https://www.um.edu.mt/library/oar/handle/123456789/146069</link>
      <description>Title: Feedback loops and bias in machine learning algorithms for predictive policing
Abstract: Predictive policing describes several emerging practices of implementing artificial&#xD;
intelligence and machine learning in police work, specifically in attempting to predict&#xD;
future crimes through algorithmic crime forecasting. These emerging practices have&#xD;
introduced many new opportunities for improved police work, but critics of predictive&#xD;
policing have raised both ethical and practical concerns. These concerns include the&#xD;
risk of feedback loops and bias. This thesis aims to contribute to this ongoing debate&#xD;
by examining how algorithmic crime forecasting tools produce bias and feedback loops&#xD;
and by exploring if it is possible to create algorithmic crime forecasting tools with&#xD;
reduced tendencies towards bias and feedback loops.&#xD;
Specifically, the focus is on the seminal and widely adopted PredPol system,&#xD;
which is based on an earthquake prediction system known as Epidemic Type&#xD;
Aftershock Sequence (ETAS). The methodology used in this studywas to replicate&#xD;
studies detailing the PredPol system, as well as studies criticising it. Based on previous&#xD;
findings by critics, a synthetic population and urn modelling was used to demonstrate&#xD;
the negative tendencies of the system.&#xD;
Based on this, an original framework was developed for evaluating&#xD;
modifications made to the algorithm by measuring the effectiveness in reducing&#xD;
feedback loop tendencies and improving fairness. This is done through metrics like&#xD;
the Predictive Accuracy Index (PAI), variations in the mean conditional intensity&#xD;
rates, λ, and the total fairness score, which evaluates the consistency of law&#xD;
enforcement attention across different demographic groups.&#xD;
To reduce the algorithm’s tendencies towards bias and feedback loops, a&#xD;
modified algorithm using rejection sampling and a fairness penalty was developed.&#xD;
While the proposed algorithmic adjustments lead to increased fairness and reduced&#xD;
feedback loop generation in predictive policing, they also introduce some trade-offs in&#xD;
predictive performance, particularly noted in the PAI values. However, the&#xD;
enhancements significantly mitigate biased policing practices and reduce the&#xD;
perpetuation of historical inequities, aligning more closely with ethical standards.
Description: M.Sc. ICT(Melit.)</description>
      <pubDate>Wed, 01 Jan 2025 00:00:00 GMT</pubDate>
      <guid isPermaLink="false">https://www.um.edu.mt/library/oar/handle/123456789/146069</guid>
      <dc:date>2025-01-01T00:00:00Z</dc:date>
    </item>
    <item>
      <title>Monitoring incomplete traces navigating uncertainty</title>
      <link>https://www.um.edu.mt/library/oar/handle/123456789/146067</link>
      <description>Title: Monitoring incomplete traces navigating uncertainty
Abstract: As software systems scale in size and complexity, manual code reviews&#xD;
become increasingly more resource intensive, underscoring the critical&#xD;
need for reliable Runtime Verification (RV) techniques. Traditional RV                                                                    approaches assume complete access to error-free execution trace, an                                                                  assumption that is untenable in real-world scenarios. This work address&#xD;
the challenges of monitoring such incomplete traces by proposing a novel&#xD;
methodology that synthesises sound and modular runtime monitors.&#xD;
Building upon detectEr, our monitors are able to handle non-consecutive&#xD;
missing events in observed traces. We demonstrate a scalable and sound&#xD;
framework that facilitates verdict generation through deterministic state&#xD;
inference, enabling monitors to achieve greater completeness on data&#xD;
restricted traces. The methodology combines a theoretically rigorous                                                                            approach with practicality in mind, utilising an automaton-guided state                                                                    regeneration technique for state inference.&#xD;
The findings from our evaluation demonstrate the efficacy of the                                                                           proposed solution across several examples. The modular monitors are                                                                  capable of navigating through data-restricted traces and producing sound,&#xD;
irrevocable verdicts. This thesis contributes to the field of RV by establishing                                                           a sound foundation for the deterministic inference of missing events&#xD;
in incomplete traces, and also provides a framework that increases the&#xD;
applicability of monitors by taking a compositional approach to monitor&#xD;
synthesis.&#xD;
By bridging a gap that is often overlooked in this field,such a solution is                                                                  required, especially in recent years, as systems are becoming exponentially&#xD;
more complex.
Description: M.Sc. ICT(Melit.)</description>
      <pubDate>Wed, 01 Jan 2025 00:00:00 GMT</pubDate>
      <guid isPermaLink="false">https://www.um.edu.mt/library/oar/handle/123456789/146067</guid>
      <dc:date>2025-01-01T00:00:00Z</dc:date>
    </item>
    <item>
      <title>Investment portfolio through evolutionary algorithm</title>
      <link>https://www.um.edu.mt/library/oar/handle/123456789/146015</link>
      <description>Title: Investment portfolio through evolutionary algorithm
Abstract: This dissertation investigates the application of evolutionary algorithms—specifically &#xD;
the Genetic Algorithm (GA) and Simulated Annealing (SA)—for portfolio optimisation &#xD;
within the S&amp;P 500, addressing the limitations of traditional models. The &#xD;
methodology uses Gower’s distance to handle mixed numerical and categorical data, &#xD;
allowing for the construction of factor-aligned portfolios based on Growth, Value, and &#xD;
Quality dimensions. The primary optimisation objective is to maximise the Sortino &#xD;
Ratio, focusing on downside-risk-adjusted returns. &#xD;
The methodology employs composite feature engineering to rank stocks across &#xD;
Growth, Value, and Quality dimensions, followed by distance-based clustering and &#xD;
anomaly detection to reveal market structures. GA and AGA are then applied with &#xD;
objective functions designed to maximise the Sortino Ratio, which emphasises &#xD;
downside-risk-adjusted returns. Hyperparameters such as population size, mutation &#xD;
rate, crossover probability, and annealing temperature schedules are tuned to &#xD;
balance exploration and exploitation. &#xD;
Empirical evaluation demonstrates that both GA and AGA generate highly diversified &#xD;
portfolios with competitive performance. The GA-optimised maximum Sortino &#xD;
portfolio achieved an annualised return of 18.95%, a Sortino Ratio of 1.0999, and a &#xD;
beta of 0.98, indicating strong returns with reduced downside risk relative to the &#xD;
market. Comparative analysis reveals that AGA converges faster and achieves &#xD;
marginally superior downside protection, validating its advantage in complex search &#xD;
landscapes. Complementary tools such as similarity maps, hierarchical clustering, &#xD;
and a diversification recommender further enhance interpretability and practical &#xD;
applicability. &#xD;
The results underscore the potential of evolutionary algorithms to construct robust, &#xD;
risk-aware investment portfolios that go beyond linear optimisation frameworks. By &#xD;
combining factor-based insights with evolutionary optimisation, this work contributes &#xD;
to the growing literature on computational finance and demonstrates actionable &#xD;
applications for institutional and retail portfolio managers.
Description: M.Sc.(Melit.)</description>
      <pubDate>Wed, 01 Jan 2025 00:00:00 GMT</pubDate>
      <guid isPermaLink="false">https://www.um.edu.mt/library/oar/handle/123456789/146015</guid>
      <dc:date>2025-01-01T00:00:00Z</dc:date>
    </item>
  </channel>
</rss>

