Data Science Commons^™

Exploring The Determinants Of Life Expectancy At Birth: Predicting And Forecasting Global Health Trends Using Statistical, Machine Learning, And Deep Learning Models., Emma Rath

ICT

Accurate life expectancy forecasting is essential for health policy planning, yet research comparing statistical, machine learning, and deep learning approaches under real-world constraints remains limited. This study evaluates ARIMA/ARIMAX, tree-based, and neural network models using Irish and global datasets, considering small samples, missing data, and COVID-19 shocks. ARIMAX with lagged socioeconomic variables outperformed LSTM and other ML/DL methods. Income-based stratification improved predictive accuracy and interpretability, with SHAP analysis highlighting GDP per capita for developed countries and school enrolment and trade indicators for developing contexts. Results provide practical guidance for policymakers and establish limits for model complexity under constrained health data.

Identifying Skills Gaps And Labour Market Trends Using Machine Learning: A Data Analysis Of Tendencies Within The Irish Online Job Market., Stephen Ray

ICT

The Irish labour market is a dynamic system characterised by overlapping and distinct demands. The skills required to maintain competitiveness within today's labour market are constantly evolving in response to the needs of corporations, non-governmental organisations, and capital. Labour market participants, fulfilling specific roles, contribute to a skill-based relationship where supply and demand operates in new and innovative ways. With the advent of the internet, the labour market has shifted towards a digital environment, leading to the rise of online job advertisement platforms that cater to geographically defined regions. These platforms attract vast numbers of users and prospective employees daily …

Forecasting Hourly Police Call For Service Volumes: A Comparative Analysis Of Statistical, Machine Learning And Neural Network Models For Operational Planning., Patrick Duggan

ICT

This study investigates hourly call volume forecasting for New Orleans police service data, comparing statistical, tree-based, and deep learning approaches. ARIMA models represent traditional methods, XGBoost and Random Forest serve as tree-based ensembles, and Gated Recurrent Units (GRUs) provide deep learning alternatives, with a naive seasonal model as a baseline. Models are evaluated using a practical, expanding time-block framework simulating real operational deployment. Results show GRUs achieve the highest accuracy (R² = 0.74, MAPE = 13.5%), with XGBoost performing similarly, while ARIMA underperforms. Additionally, a Random Forest model offers interpretability, identifying key factors that drive forecasting performance, providing actionable insights …

Forecasting Hourly Police Call For Service Volumes: A Comparative Analysis Of Statistical, Machine Learning And Neural Network Models For Operational Planning., Patrick Duggan

ICT

Accurate demand forecasting is critical in operational settings where resource allocation and planning decisions depend on anticipated service volumes. Transactional systems that capture timestamped records provide valuable data sources for developing demand forecasts. This study examines hourly call volume forecasting using New Orleans police calls for service data, comparing the performance of statistical models, tree-based methods, and recurrent neural networks.

The research evaluates four primary modelling approaches: ARIMA models representing traditional statistical methods, XGBoost and Random Forest as a tree-based ensemble technique, and Gated Recurrent Units (GRUs) as deep learning alternatives. A naive seasonal model serves as the baseline benchmark. …

Predictive Analysis Of Atlantic Meridional Overturning Circulation Amoc And Its Climatic Implications On Ireland Using Advanced Machine Learning Techniques., Corey Whyte

ICT

The Atlantic Meridional Overturning Circulation (AMOC) plays a key role in regulating climate across the North Atlantic, including Ireland. This study analyses AMOC variability using RAPID, ODYSSEA, and MÉRA datasets to examine relationships between subsurface ocean conditions, sea surface temperatures, and air temperature. ARIMA/SARIMA models identify trends and correlations, while a Long Short-Term Memory (LSTM) neural network forecasts future climate changes. Results show rising sea surface temperatures, moderate links between salinity and temperature, and seasonal air temperature patterns with a slight upward trend. The LSTM model outperforms traditional statistical approaches in capturing complex patterns, highlighting the value of machine learning …

Predicting S&P Corporate Credit Ratings Using Financial Ratios And Machine Learning: An Analysis Of European Non-Financial Companies., Gabriele Frattaroli

ICT

This study investigates the prediction of multi-class S&P corporate credit ratings for European non-financial firms from 2010 to 2024 using a machine learning framework grounded in financial fundamentals. To ensure robustness and generalizability, the analysis excluded the Year variable, which was identified as a source of data leakage. After this correction, non-linear ensemble models demonstrated a clear advantage over linear baselines. The top-performing Random Forest model achieved a weighted F1-score of approximately 0.60, more than doubling the performance of the Logistic Regression benchmark used as a baseline (0.26), with most misclassifications concentrated in adjacent rating categories. This indicates that while …

Predicting Monthly Weather Anomalies In Ireland: A Comparative Study Of Machine Learning And Deep Learning Models., Fiona Behan

ICT

Abstract

This study investigates the application of statistical, machine learning, and deep learning methods to detect monthly weather anomalies in Ireland between 1960-2024. Climate variability is intensifying globally, increasing the urgency for accurate detection of unusual weather events. Using publicly available data from Met Eireann provided by the CSO PxStat Open Data Portal, the research applied comprehensive preprocessing including Bayesian Ridge iterative imputation, temporal and seasonal feature engineering, and ensemble statistical anomaly labelling based on z-score, interquartile range, and rolling residual analysis. Four models were developed and evaluated: Isolation Forest, XGBoost, Long Short Term Memory networks, and Gated Recurrent Units, …

Optimizing Lstm Neural Network For Multimodal Multivariate Footfall Prediction., Aws Al Adhami

ICT

Accurate forecasting of pedestrian activity is important for smart city planning, retail analytics, and public service management. This study presents a forecasting framework using Long Short-Term Memory (LSTM) neural networks that integrates multiple data sources, including weather, calendar variables, and engineered temporal features. The model development follows a structured pipeline from baseline univariate models to a fully optimized multivariate model using feature engineering and hyperparameter tuning techniques such as Hyperband and Bayesian optimization. The final model achieved a Mean Absolute Error of 24.05 and explained 97.10% of the variance in unseen data, effectively capturing both regular patterns and short-term fluctuations …

Neural Networks Activation Functions And Hybrid Activations Functions Accuracy And Loss Comparison On Small Dataset Against Large Datasets For Classification Problems., Antonio Felipe Cora Martins

ICT

Even on the era of Big Data, small datasets are the reality of many companies and sectors. Many datasets in rare disease diagnosis, custom manufacturing, military sciences, bioengineering, and disaster events are commonly limited in size, making machine learning predictive modelling difficult. Being the Activation Function choice crucial for Neural Networks learning, it raises the question of their effectiveness in such scenarios. This study compares five standard (single) activation functions (Sigmoid, Tanh, ReLU, Leaky ReLU, ELU) and two hybrid variants (one a mix of ReLU plus Tanh and a Learnable Activation Function with a trainable weight (alpha) that balances ReLU …

Leaf Classification Using Convolutional Neural Networks And Vision Transformers, Louis Wilkie

ICT

This study investigates deep learning techniques for plant image classification using four publicly available datasets: Folio, Flavia, Swedish, and Sugarcane. Three modelling approaches—Convolutional Neural Networks (CNN), Vision Transformers (ViT), and hybrid CNN-ViT ensembles—are evaluated. Models leverage PyTorch implementations and pretrained weights from the Hugging Face hub, with hyperparameter tuning applied to optimize accuracy. Transfer learning with pretrained ViT models achieved the highest performance, attaining 100% accuracy on Folio, Flavia, and Swedish datasets, and 94.07% on Sugarcane. CNN and CNN-ViT models showed comparable results, but pretrained CNN-ViTs were more resource-efficient. The findings highlight the importance of evaluating multiple pretrained architectures to …

Investigating The Effectiveness Of Traditional Vs Hybrid Time Series Models In Operational Planning., Cristina Priolo

ICT

This study conducts a systematic comparison of traditional and hybrid time series forecasting models for operational planning in a SaaS company, using large multichannel, multilanguage datasets aggregated at six-hour intervals. Classical models (ARIMA, SARIMA, ETS) are evaluated against hybrid models (ARIMA-ANN, SARIMA-ANN, ETS-ANN) to assess their ability to capture both linear and nonlinear patterns. Forecasts over a three-month horizon were evaluated using RMSE, MAE, and MAPE, with hyperparameter optimization applied to all models. Results show that hybrid models, particularly ARIMA-ANN and SARIMA-ANN, outperform traditional models in predicting volatile and high-volume data, while traditional models remain competitive for stable datasets. These …

Graph Neural Networks Powered Scientific Paper Recommendation, Junhao Shen

Computer Science and Engineering Theses and Dissertations

Scientific paper recommendation systems aim to help researchers discover relevant papers amidst the vast and ever-growing body of literature. With the exponential yearly increase in scientific publications, the demand for effective paper recommendation solutions has become both critical and increasingly challenging. In recent years, deep learning techniques have revolutionized recommender systems, and scientific paper recommendations have naturally integrated these advancements. In this dissertation, we address these challenges through three progressive contributions.

First, we enhance traditional content-based methods using Graph Neural Networks (GNNs) by introducing a Graph Convolutional Network-strengthened Topic Modeling (GCN-TM) approach. This method improves upon conventional topic modeling techniques …

The Quantitative Analysis And Visualization Of Nfl Passing Routes, Sandeep Chitturi

Computer Science and Computer Engineering Undergraduate Honors Theses

The strategic planning of offensive passing plays in the NFL incorporates numerous variables, including defensive coverages, player positioning, historical data, etc. This project develops an application using an analytical framework and an interactive model to simulate and visualize an NFL offense's passing strategy under varying conditions. Using R-programming and data management, the model dynamically represents potential passing routes in response to different defensive schemes. The system architecture integrates data from historical NFL league years to generate quantified route scores through designed mathematical equations. This allows for the prediction of potential passing routes for offensive skill players in response to the …

A Holistic And Collaborative Behavioral Health Detection Framework Using Sensitive Police Narratives, Martin Keagan Wynne Brown

Dissertations

Identifying behavioral health is paramount for law enforcement officers to provide appropriate follow-up community care. In the current practice, law enforcement offices manually identify these behavioral health cases to allow the designation of the relevant follow-up resources. Police reports generated by officers' response to 911 calls remain an untapped resource for identifying such incidents. Therefore, we advocate for the incorporation of manual annotations from experts, natural language processing (NLP), active learning, advanced machine learning, and ensemble techniques to detect behavioral health cases within police reports. In this dissertation, we develop tools and frameworks to automatically detect behavioral health cases from …

Comparison Of Classification Methodologies Using Convolutional Neural Networks In A Dataset Of Plant Leaf Diseases., Ruairi O’Donohoe

ICT

This project investigates the impact of classification methodology selection on the performance of four Convolutional Neural Network (CNN) models applied to a multi-label image dataset. The dataset consists of plant leaf images with one or more diseases. Two classification methodologies—multi-label and multi-class—are compared based on their model performance metrics. It was hypothesised that multi-label classification would perform better, but the results show that although multi-label models performed better for Loss and Accuracy metrics, they underperformed in terms of the F1 score, which is considered a more appropriate metric for this task. This surprising result refutes the initial hypothesis. Transfer learning …

Supply Chain Optimisation With Machine Learning And Neural Networks: Applications To Demand Planning, Supply Planning, And Inventory Planning., Laurence Cully

ICT

This thesis explores the impact of machine learning (ML) on supply chain planning, particularly in demand forecasting, supply planning, and inventory optimisation. By analysing literature on supply chain management, data flow, and the intersection of ML and competitive advantage, the author contextualises the research within a globalised market's demands. Case studies, interviews with industry professionals, and raw data collection provide empirical support for evaluating the research objectives and documenting the integration of ML in supply chain processes.

The findings reveal that optimised ML models, particularly those using model stacking (autoregressors, GRUs, and Random Forests), significantly outperform traditional demand forecasting methods, …

Evaluation And Implementation Of Machine Learning Models To Predict Customer Churn In The Telecommunications Sector., Stephen Hasson

ICT

This research addresses customer churn in the Telecom industry by utilizing Machine Learning (ML) models to predict customers at risk of leaving and provide data-driven retention strategies. The study highlights the effectiveness of ML, particularly in churn prediction, while noting the need for further exploration into the ethical implications of AI, such as potential biases towards vulnerable groups. Using the CRISP-DM framework, the study develops and compares three Supervised Learning (SL) models: Random Forests (RF), LightGBM (LGBM), and XGBoost (XGB), incorporating class resampling techniques to manage data imbalance.

The findings identified five key features as the most significant predictors of …

Statistical And Machine Learning Techniques For Predicting Solar Power Generation In A Microgrid., Conor Dillon

ICT

This study investigates statistical and machine learning models for forecasting solar power generation in microgrids, focusing on the solar installation at Powell-Focht Bioengineering Hall, UC San Diego. Accurate predictions are critical due to the variability of solar energy, aiming to optimise microgrid operations and solar power efficiency. The research compares the performance of SARIMAX, LSTM, Random Forest, and ANN models using meteorological and solar power time series data. It finds that current meteorological inputs, especially solar radiation, enhance short-term forecasting accuracy over reliance on historical patterns.

The Random Forest Auto Regressor (RFAR) outperformed other models in 10-day-ahead solar power forecasting, …

Responsible Natural Language Processing To Aid Employee Performance Reviews., Grace Rubinger

ICT

This research explores the use of Natural Language Processing (NLP) techniques in assessing evaluators' written appraisals during Employee Performance Reviews (EPRs), aiming to address biases inherent in traditional methods. By integrating Responsible Artificial Intelligence (AI) and foundational Large Language Models (LLMs), the study seeks to enhance the objectivity, fairness, and ethical transparency of performance evaluations. It highlights the potential of AI systems to ensure comprehensive assessments while promoting trust, ethical standards, and employee retention.

The research also aims to advance the field of AI Ethics in practical Human Resources Management (HRM) applications, particularly through NLP-driven tools. These tools are designed …

Using Machine Learning To Identify Hate Speech And Offending Language On Twitter., Mayara Lorens, Thayene Lorens

ICT

This project focuses on applying Machine Learning (ML) techniques to detect hate speech and offensive language on Twitter, addressing ethical concerns like cyberbullying and fostering a safer online environment. The topic is chosen for its societal significance and business relevance, as hostile online behaviour negatively impacts user experiences and platform credibility.

To achieve this, the study implements four distinct ML models to develop an automated system capable of identifying and categorising content as offensive, non-offensive, or neutral. The system aims to contribute to mitigating harmful interactions on social media and improving user safety by effectively classifying potentially problematic content.

The …

Using Unsupervised Learning Methods In Extracting Features For Classifying Rice Varieties From Rice Grains Images., Kevin Anthony Martinez

ICT

Rice, a staple food for nearly half of the global population, requires accurate classification of its varieties to ensure food quality, support agricultural trade, and enhance yield optimisation. Traditional manual classification methods are time-intensive and error-prone, prompting this study's exploration of unsupervised learning for feature extraction from rice grain images. The research tested classifiers on 75,000 rice samples across five classes, with 15,000 samples per class.

The study's DCGAN-CNN model achieved the highest classification accuracy of 99.67%. However, the PCA-CNN model underperformed, with only 20% accuracy, due to implementation errors. Recommendations for improvement include optimising model parameters such as learning …

Applying Neural Networks To Predict Factors Affecting Harmful Algal Blooms For Timely Alerting And Implementing Preventive Measures In Ireland's Marine Ecosystem., Nikolai Potapov

ICT

This study applies neural networks to predict harmful algal blooms (HABs) along the Irish coast, addressing ecological, health, and economic risks. Using primary interviews and secondary data on HAB species like Alexandrium and Karenia mikimotoi, the research incorporated Exploratory Data Analysis and tested three neural models: LSTM, Ensemble Stacking LSTM, and CNN-LSTM. Key factors influencing HABs, such as sea surface temperature and euphotic zone depth, were identified.

Results demonstrate the potential of neural networks to improve HAB prediction and monitoring, despite limitations. Future work aims to enhance model accuracy and integrate them into HAB warning systems.

Assessment Of The Impact Of Various Feature Extraction Techniques On The Effectiveness Of Music Genre Classification In Neural Network Models., Sabhdh Grace

ICT

This research focuses on Music Genre Classification (MGC) using Convolutional Neural Networks (CNNs) and various datasets, including raw audio files (WAV) and extracted features such as Mel Spectrograms (MS), Mel-Frequency Cepstral Coefficients (MFCC), and Chroma Features (CF). The study employs Explanatory Sequential Mixed Methods (ESMM), combining qualitative research and experimental analysis to explore different model inputs and their performance. Several CNN-based models, including 2D CNN, 2D CNN-LSTM, 1D CNN, and 1D CNN-LSTM, were tested. However, the models generally underperformed, with most achieving accuracy of 10% or lower, and the best model (raw audio 1D CNN) reaching only 20%. The research …

Deep Learning Model Compression For Resource-Constrained Environments., Stephen Burke

ICT

This study examines the effects of three Deep Neural Network compression techniques—Quantisation, Pruning, and Weight Sharing/Clustering—on CNN and ANN models trained for image classification tasks. The models were tested on the CIFAR-10 dataset for multiclass classification and a binary classification task using a dataset derived from COCO. The best validation accuracy achieved was 74.7% with a CNN on CIFAR-10 and 53% with the best ANN. On the COCO dataset, a modified CIFAR-10 CNN model achieved 75%. The models were compressed using the three techniques and benchmarked on a ThinkPad laptop and Raspberry Pi 3B+ based on metrics relevant for resource-constrained …

Data Analysis Of Twitter’S Nasdaq100 Sentiments And Topics As Indicators For News Articles Retrieval: Fine-Tuning Roberta And Rag., Kagan Timur

ICT

This study investigates the combination of sentiment analysis using the VADER lexicon and semantic analysis through Latent Dirichlet Allocation (LDA) to identify real-life events, focusing on Twitter datasets. The research shows that while sentiment analysis alone may be insufficient, combining it with semantic analysis improves the process, particularly for identifying relevant news articles and understanding brand perception on social media. The study also fine-tunes the RoBERTa model for question-answering tasks, yielding significant improvements in the SQuAD evaluation metric. The exact match (EM) score rose dramatically from 2.06% to 62%, and the F1 score improved from 9.41% to 65%. A retrieval …

Development And Optimisation Of Convolutional Neural Networks (Cnns) To Predict The Nutrition And Sustainability Scores Of Foods From Crowd Sourced Images., Cormac Mcelhinney

ICT

This research explores the use of Convolutional Neural Networks (CNNs) for the automated classification and profiling of food products based on publicly sourced data. With the vast array of food products available worldwide and the complexity of labelling regulations, food business operators face challenges in ensuring compliance, while regulators struggle to verify adherence. This study addresses the need for efficient and accurate methods for food classification and eco/nutritional profiling. It begins with a comprehensive literature review on the application of CNNs in food product classification, followed by the collection of a large-scale dataset from Open Food Facts. A CNN architecture …

Evaluating The Performance Of Different Long Short-Term Memory Networks (Lstm’S) On Financial Timeseries Data Using Mean Squared Error In Order To Identify The Optimum Lstm Variant For Regression Performance On Financial Timeseries Data., Patrick O’ Connor

ICT

This study explores the use of Long Short Term Memory (LSTM) networks, a variant of Recurrent Neural Networks (RNNs), in the context of financial forecasting, specifically oil price prediction. The research follows the CRISP-DM (Cross-Industry Standard Process for Data Mining) methodology and tests six different LSTM variants. The models are evaluated based on Mean Squared Error (MSE), aiming to determine the optimal parameter settings for each LSTM type. Among the variants tested, the Gated Recurrent Unit (GRU) emerged as the highest performer, achieving an MSE of 0.100. This was surprising, as simpler variants outperformed more complex ones, suggesting that simpler …

Ml Predictive Model For Earthquakes Integrating Mass, Distance, Gravity, And Magnitude., Aadarsh Kushwaha

ICT

This research investigates the application of machine learning regression models to improve earthquake prediction by integrating geophysical and astronomical factors such as Earth-Moon gravitational forces, their varying distances, and localized gravity fluctuations. Using data from 2011 to 2024, sourced from the US Geological Survey (USGS) and web scraping, the study tested models across four dataset proportions (25%, 50%, 80%, and 100%) with a 70:30 train-test split. The XGBRegressor model emerged as the best performer, achieving an R² score of 0.8706 on training data and 0.8632 on test data, along with a Mean Squared Error (MSE) of 0.1114 and Mean Absolute …

Maize Crop Pests And Diseases Classification Using Hybrid Models., Diana Flora Namaemba

ICT

This research focuses on improving the detection and classification of maize crop pests and diseases to enhance agricultural yield and food security. A dataset comprising 5389 images of maize conditions (healthy, pest-affected, and disease-affected) across seven classes was used. The images underwent preprocessing, including resizing to 299x299, class balancing using augmentation techniques, and noise reduction with Gaussian filtering.

Feature extraction utilised EfficientNetB0 and InceptionV3 architectures, with PCA employed for feature selection. Classification was conducted using a Support Vector Machine (SVM) with a One-vs-One strategy, alongside a baseline 2D CNN model. Data engineering included label encoding, standardisation, and an 80:10:10 train-test-validation …

An Empirical Study Of Machine Learning Techniques For Accurate Stock Price Forecasting, Daniel Paliulis, Hari Patchigolla

Honors Scholar Theses

This paper presents a comprehensive approach to predicting future stock prices of companies using machine learning and time series analysis. The research problem is centered around addressing the complexity and emotion-driven nature of stock investment decisions. To create an objective determinant in stock decisions, we propose a machine learning model utilizing time series data from major companies, including Amazon, Apple, Google, Nvidia, Meta, Tesla, Salesforce, Intel, and Microsoft. We explore the use of Long Short-Term Memory (LSTM) neural networks, to capture the temporal dynamics of stock prices. These models are designed to process sequential data, maintaining short term and long …