The University of Queensland

Publications

PhD and MPhil Supervision

Possible Research Projects

Note for students: The possible research projects listed on this page may not be comprehensive or up to date. Always feel free to contact the staff for more information, and also with your own research ideas.

• Predicting the solutions of optimisation models with Machine Learning

  Prescriptive analytics or operations research is a well-studied field that deals with finding optimal solutions for complex problems. There is a vast literature on prescriptive analytics that use linear programming, non-linear programming, mixed integer programming, and constrained programming techniques to solve optimisation problems. Recently, there has been an interest in using machine learning to solve these problems. Although machine learning models may not give the optimal solutions, they are significantly faster and capable of generating near-optimal solutions.

  The aim of this project is to use advanced machine learning and deep learning models to solve constrained optimisation problems.

• Forecasting and then optimising decisions

  Predictive and prescriptive analytics are two prominent techniques capable of addressing numerous real-world challenges. The "predict and optimise" paradigm proves invaluable in scenarios demanding dynamic decision-making. For instance, one might aim to predict product demand to fine-tune production planning, or forecast electricity demand to optimally schedule machine operations. This approach has manifold applications across sectors like finance, retail, manufacturing, and energy. Within this context, predictions serve as inputs to optimisation models. While heightened prediction accuracy often bolsters optimisation, it doesn't always directly lead to enhanced results. The core challenge we seek to address is the seamless integration of these two phases to craft an end-to-end model focused on decision optimisation. Throughout this process, you will hone machine and deep learning models that consider final decisions in their forecasting efforts.

• Clustering applications to improve the forecasting accuracy of models

  Developing a forecasting model for a set of similar series together has been shown to improve the forecasting accuracy as opposed to having one model for all. This project aims to look at the clustering methods and how they can be used to improve the forecasting accuracy of energy demand. We will aim to innovatively cluster the most similar days to develop and improve the accuracy of models.

• Hierarchical forecasting with constrained optimisation

  Forecasting is the basis for many managerial decisions, such as inventory control, budgeting, and staff planning. Hierarchical time series comprise several time series that can be organized at hierarchical levels based on various features, such as geographical regions or product categories. Hierarchical time series forecasting is essential in numerous situations, as time series are often hierarchical in nature. Overall, this is a significant problem, and your work will further advance both the theory and practice of forecasting, developing practical skills. In this project, you will create constrained optimization models for forecasting time series across different levels according to our needs. This is empirical research and can be implemented using open-source data from the M5 forecasting competition.

• Hierarchical Forecasting with Machine Learning- Graph Neuwral Network

  Hierarchical Forecasting has found many applications in real-world. Hierarchy refers to multiple time series that have a natural connection, e.g, across locations, or time. This project aims to use graph Neural Network in the setting of hierarchical time series, to develop an end-to-end algorithm that is able to forecast the entire series in an optimal way.

• Forecasting outliers

  Outlier detection is an important problem in many fields including in time series forecasting. Applications include detecting large spikes in transactions, or security breaches. There are some standard techniques that can be used for the early detection of outliers, e.g. extreme value theory. In this project, we aim to connect forecasting and outlier detection by leveraging the techniques in both literature to not only detect outliers but be able to forecast them with high precision.

• Optimisation of frameworks for development and operation of automated sound classification in real-world environmental monitoring

  Effective methods for sound classification are widely published, but these works often reference highly curated datasets or are applied to tightly controlled scenarios. Accurate sound classification in real-world environments are confounded by variability in signal to noise ratios and variability in the characteristics of noise sources. This work seeks to explore the influence that data representations (i.e. data engineering) and construction of training algorithms may have on the performance of environmental noise classification. An existing classification framework and training dataset are available for the purposes of baselining ‘existing’ performance.

  Expected Outcomes

  Improved understanding of influence that algorithms and data representations have on performance of noise classification problems.

  This is an industry-supported project. The interested student will work closely with Advitech.

  Suitable for Honours students.

• How forecasts uncerstainity can impact the final decisions?

  Probabilistic forecasting associates a probability of occurrence with the predicted values, making it a useful technique for determining decisions based on the level of risk one can take. It is a powerful technique that unlike point forecast gives you a complete view of the future unknown values. In this project, we aim to use probabilistic forecasting to predict the demand for products/services and accordingly determine a better decision whatever that may be, e.g., inventory of product, or optimal allocation of resources. We investigate the association between these two in the context of Energy and supply chain inventory.

• Machine learning applications in forecasting and anomaly detection in cybersecurity

  This research project explores the application of machine learning techniques in the fields of cybersecurity forecasting and anomaly detection. With the ever-growing sophistication of cyber threats, traditional security measures are often insufficient to protect systems and networks effectively. By leveraging machine learning algorithms, this study aims to develop accurate and efficient models for predicting cyber attacks and identifying anomalous behavior. The project involves analyzing large datasets of historical cybersecurity incidents, extracting relevant features, and training models to recognize patterns indicative of malicious activities. The findings of this research have the potential to enhance the proactive cybersecurity measures and bolster defense mechanisms against evolving cyber threats.