https://www.um.edu.mt/library/oar/handle/123456789/140233 2026-04-11T14:19:30Z 2026-04-11T14:19:30Z https://www.um.edu.mt/library/oar/handle/123456789/141020 2025-11-10T07:33:22Z 2025-01-01T00:00:00Z

Title: Identifying optimal investment strategies with deep reinforcement learning Abstract: The rise of fully automated trading systems has transformed global financial markets, placing greater emphasis on intelligent data-driven decision making. This thesis explores the development of optimal investment strategies using Deep Reinforcement Learning (DRL), with a particular focus on the Proximal Policy Optimisation (PPO) algorithm. Historical closing price data from a diversified portfolio of seven technology stocks was collected, processed and combined with market indicators to form the model inputs. A supervised learning baseline was first established using a Multilayer Perceptron (MLP) to provide a performance benchmark. Subsequently, DRL agents that incorporate different neural network architectures, MLPs, Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN), were implemented within a custom PPO framework designed for multiple stock portfolio management. Each model was evaluated using two state representations: normalised closing prices and normalised engineered market features, enabling a comparison of model performance under varying input dimensions. The evaluation was carried out using Monte Carlo rollouts in a custom simulated trading environment using 2023 test data. The PPO agent with an MLP architecture and engineered features achieved the most stable returns, averaging a gain of 152%, while the CNN-based agent with closing price-only input reached a maximum return of 265% but with a higher volatility. These results suggest that, when the models are appropriately structured and trained, DRL agents can outperform both traditional supervised learning approaches and passive strategies in simulated markets, offering a promising foundation for further research into adaptive algorithmic portfolio optimisation. Description: B.Eng. (Hons)(Melit.)

2025-01-01T00:00:00Z https://www.um.edu.mt/library/oar/handle/123456789/140563 2025-10-24T12:38:01Z 2025-01-01T00:00:00Z

Title: Towards a secure urban traffic network Abstract: Traditional Intelligent Transport Systems (ITS) face several critical limitations such as single points of failure due to centralized control and data. These limitations result in ITS solutions that are often inefficient and unreliable in meeting modern transportation demands. Recent research highlights blockchain technology as a promising solution to these challenges. By decentralizing control, securing data immutably, and enabling transparent, distributed decision-making, blockchain can strengthen ITS against failures and manipulation while improving responsiveness to dynamic traffic conditions. This dissertation presents the design, development, and implementation of a decentralized traffic management system that integrates blockchain technology with a traffic simulation. A locally hosted Ethereum Virtual Machine (EVM)-compatible blockchain (via Ganache) is connected to the Aimsun Next 23 simulation platform using Python APIs, enabling real-time, bidirectional communication between the simulation and the blockchain network. The system uses three smart contracts: one to log vehicle counts using event emissions, another to manage actuated traffic light logic based on real-time traffic data and another to establish priority to emergency vehicles. This decentralized approach enables tamperproof data logging, distributed control, and programmable traffic responses. It also supports the logging and management of emergency events, such as simulated lane closures, without relying on centralized control. Four simulations were conducted to evaluate the system’s functionality, ranging from basic traffic data logging to full actuated signal control under emergency conditions. The results demonstrate that integrating blockchain technology into ITS frameworks is favourable, leading to several benefits such as transparency, resilience, and dynamic traffic management capabilities. The dissertation concludes that the prototype meets its objectives. Future work could explore deployment on scalable platforms like Polygon and the application of this work to other traffic networks. Description: B.Eng. (Hons)(Melit.)

2025-01-01T00:00:00Z https://www.um.edu.mt/library/oar/handle/123456789/140561 2025-10-24T12:35:19Z 2025-01-01T00:00:00Z

Title: Stock market prediction using ensemble learning methods Abstract: The stock market is complex and stochastic, which presents a significant challenge to accurately predict movements and prices. This study focuses on leveraging decision tree ensemble methods, specifically Random Forests and Extreme Gradient Boosting (XGBoost), for predicting future stock market movements. The primary focus of this study is to extract valuable stock market information whilst making the necessary changes to establish the best balance between investment risk and portfolio returns. The Random Forest and XGBoost models were trained and evaluated using historical stock data from five leading technology companies, namely Apple, Microsoft, Google, Tesla and Amazon. The results proved that both models were capable of delivering a reliable performance, generating substantial annual returns. The Random Forest model achieved a mean annual return of 47.8%, whilst the XGBoost model showed slightly lower performance, with a mean annual return of 32.0%. However, Random Forests achieved a maximum annual return of 85.2%, whereas XGBoost reached a higher maximum of 101.3%. These findings also revealed that Random Forests produced more consistent and conservative results, while the XGBoost model was more volatile and risky, occasionally achieving higher returns but with less stability. Overall, the results demonstrate that both ensemble models, Random Forests and XGBoost, were able to generalize effectively across unseen data, capturing informative patterns and relationships within the stock market data. Description: B.Eng. (Hons)(Melit.)

2025-01-01T00:00:00Z https://www.um.edu.mt/library/oar/handle/123456789/140558 2025-10-24T12:31:13Z 2025-01-01T00:00:00Z

Title: Stable self-levelling control of a Stewart platform Abstract: As the need for stability in motion continues to grow across various industries, there has been a keen interest in self-levelling platforms. These are mechatronic systems designed to automatically maintain a stable and level orientation, even when the surface they rest on is tilted or disturbed. The project focuses on a specific type of parallel manipulator called a Stewart platform, which is constructed using six linear actuators connected between a fixed base plate and a movable top plate. The movement of the top plate is controlled by adjusting the lengths of these actuators. The main objective of this dissertation is to design and implement a controller that is able to self-level the top plate of the Stewart platform without causing the system to become unstable, even in the presence of unpredictable forces or disturbances. This must be achieved while also managing the complex and highly non-linear behaviour that characterises the platform’s dynamics. This project presents the complete design and implementation of a self-levelling Stewart platform. It includes the development of a mathematical model of the platform’s dynamics using the Lagrangian formulation, followed by its application to a physically constructed system. A two-stage control strategy is implemented to achieve regulatory control and trajectory tracking, together with the self-levelling functionality based on a novel controller concept. The design and implementation of both the simulation models and the physical setup are detailed, encompassing the actuator control system, the inertial measurement system, hardware interfacing, and other essential aspects of the overall system. Extensive testing is carried out on both the simulation models and the physically implemented system, with results demonstrating stable performance and a high degree of self-levelling effectiveness for both user-defined poses and trajectories. Description: B.Eng. (Hons)(Melit.)

2025-01-01T00:00:00Z