Dr Mahdi Abolghasemi

Lecturer

Mathematics
Faculty of Science

Overview

Mahdi has a multi-disciplinary background in Engineering, Machine Learning, Business, and Statistics. Mahdi's research interests/expertise are in time series forecasting, predictive analytics, data science, and machine learning with applications in the supply chain, and renewable energies.

Research Interests

  • Predictive analytics
  • Time series forecasting (sales/demand forecasting, energy forecasting, price forecasting, traffic forecasting)
  • Applied Machine Learning
  • Optimisatoin
  • Data Science

Research Impacts

I am passionate about applied research, i.e, research with applications in solving real-world problems. I have been fortunate enough to work on several industry research projects during my working experience in the industry, and academic journey in PhD and postdoc. My research has been used in several sectors including the food and beverage supply chain, renewable energy, automotive, mining and construction. I have developed and optimised several decision-making models that are currently in use by organisations in Australia and overseas.

Publications

View all Publications

Available Projects

  • 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 optimisation problems.

  • Point of sales (POS) is the data that is recorded at the retailer level when consumers purchase the products. POS data is becoming increasingly popular for companies to predict their sales. In a supply chain, the POS data are often used by retailers to predict their sales, however, manufacturers and suppliers have not benefited enough from POS data. Retailers place their orders to suppliers as they predict their sales.

    The goal is to predict the supplier's demand. We can use either the retailer's orders or POS data to forecast suppliers' demand. The idea is to use Granger's causality​​ to forecast ​one time series​ (suppliers demand)​ from ​another​ time series​ (POS data). This is an empirical study and we will be using the real sales time series​​ of a food company.

  • M6 is the latest forecasting competition of the famous M-competition series. It is a well-respected competition in the forecasting community that attracts many academics and practitioners from all over the world. The goals of M6 competition are to develop forecasting and an optimisation model that together predict and optimise the portfolio using real-world live data of 100 stocks and ETFs. The competition runs from Feb 2022-Feb 2023, and participants are expected to submit their forecast and portfolio plan for all stocks/ETFs for each month on a rolling basis. The competition provides an excellent opportunity to learn and test state-of-the-art models in forecasting and optimisation. There are over 300,000 USD prizes in cash and a unique opportunity to write an academic paper upon some success.

    Reference: https://m6competition.com/

View all Available Projects

Publications

Journal Article

Conference Publication

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.

  • 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 optimisation problems.

  • Point of sales (POS) is the data that is recorded at the retailer level when consumers purchase the products. POS data is becoming increasingly popular for companies to predict their sales. In a supply chain, the POS data are often used by retailers to predict their sales, however, manufacturers and suppliers have not benefited enough from POS data. Retailers place their orders to suppliers as they predict their sales.

    The goal is to predict the supplier's demand. We can use either the retailer's orders or POS data to forecast suppliers' demand. The idea is to use Granger's causality​​ to forecast ​one time series​ (suppliers demand)​ from ​another​ time series​ (POS data). This is an empirical study and we will be using the real sales time series​​ of a food company.

  • M6 is the latest forecasting competition of the famous M-competition series. It is a well-respected competition in the forecasting community that attracts many academics and practitioners from all over the world. The goals of M6 competition are to develop forecasting and an optimisation model that together predict and optimise the portfolio using real-world live data of 100 stocks and ETFs. The competition runs from Feb 2022-Feb 2023, and participants are expected to submit their forecast and portfolio plan for all stocks/ETFs for each month on a rolling basis. The competition provides an excellent opportunity to learn and test state-of-the-art models in forecasting and optimisation. There are over 300,000 USD prizes in cash and a unique opportunity to write an academic paper upon some success.

    Reference: https://m6competition.com/

  • Forecasting is the base for a lot of managerial decisions such as inventory control, budget and staff planning, etc. Hierarchical time series are several time series that can be organised in hierarchical levels with respect to different features such as geographical regions, product category, etc. Hierarchical time series forecasting is needed in many situations as often time series are hierarchical in nature. Overall, this is an important problem, and your work will continue to advance the theory and practice of forecasting and develop highly practical skills.

    There are several projects available at the moment, all of which can be implemented on open source data ofM5 forecasting competition. The data is publicly available on https://www.kaggle.com/c/m5-forecasting-accuracy.There is an opportunity to scope the research in three directions:

    1. Developing deep learning models that can outperform conventional methods.
    2. Developing probabilistic forecasting models and optimising them using Monte Carlo.
    3. Developing optimisation models such as mixed-integer linear programming for optimising the forecasts across different levels of hierarchy.
  • Predictive and prescriptive analytics are two widely used techniques that together can solve many real-world problems. The paradigm of "predict and optimise" is useful when we are dealing with dynamic decision-making problems. For example, you may be interested to predict the demand for products and accordingly optimise your production planning, or you may be interested in predicting the electricity demand and optimise your scheduling for machines. There are many other examples in finance, retail, manufacturing, and energy domain where this paradigm would be useful. In this class of problems, the prediction will be used as n input to the optimisation model and often a better accuracy leads to better optimisation. However, the production accuracy will not directly translate to improvement in optimisation. The problem that we would like to answer is how we can integrate these two phases and develop an end-to-end model that can optimise the decisions.