Webinar: Learning from imbalanced data using improved data pre-processing procedures and machine learning techniques

#CIS_Chapter #CIS_Germany #computational-intelligence #classification-algorithms #machine-learning #radial-basis-function-networks #methods #imbalanced-data #overfitting #oversampling
Share

The challenge of learning from imbalanced data is well-recognised, as standard classification algorithms are often found to struggle with minority instances, which may represent rare but critical events. Improvements have been developed in both data pre-processing and Radial Basis Function Networks (RBFNs). Feature selection has been enhanced by storing and combining multiple candidate solutions, reducing overfitting and more reliably identifying relevant features. Oversampling has been refined to determine neighbourhoods and the number of synthetic instances automatically, simplifying its use. RBFN classifiers have been designed incrementally, with knowledge from simpler networks guiding the construction of more complex ones. These approaches have been shown to improve minority class recognition and overall performance, providing practical methods for tackling imbalanced data in real-world problems.


  Date and Time

  Location

  Hosts

  Registration


  • Add_To_Calendar_icon Add Event to Calendar

Loading virtual attendance info...

  • Contact Event Host
  • Starts 19 November 2025 11:00 PM UTC
  • Ends 08 December 2025 04:00 AM UTC
  • No Admission Charge

  Speakers

Dr. Mario Dudjak

Topic:

Learning from imbalanced data using improved data pre-processing procedures and machine learning techniques

The challenge of learning from imbalanced data is well recognised, as standard classification algorithms are often found to struggle with minority instances, which may represent rare but critical events. Improvements have been developed in both data pre-processing and radial basis function networks. Feature selection has been enhanced by storing and combining multiple candidate solutions, reducing overfitting and more reliably identifying relevant features. Oversampling has been refined to determine neighbourhoods and the number of synthetic instances automatically, simplifying its use. RBFN classifiers have been designed incrementally, with knowledge from simpler networks guiding the construction of more complex ones. These approaches have been shown to improve minority class recognition and overall performance, providing practical methods for tackling imbalanced data in real-world problems.

Biography:

Dr. Mario Dudjak is a postdoctoral researcher at the University of J. J. Strossmayer in Osijek, specialising in machine learning and software engineering. He earned his BSc, MSc and PhD in Computer Science from the Faculty of Electrical Engineering, Computer Science and Information Technology Osijek. During his studies, he participated in competitions such as IEEE Extreme and IEEE MADC and received several academic awards for outstanding achievement.

Mario has worked on numerous research and professional projects involving spatial data processing, high performance computing, optimisation and computer vision, contributing as an expert in machine learning and data science. He has published dozens of papers in scientific journals and international conferences, including a notable first author publication in Expert Systems with Applications (IF 8.665). He also teaches several courses in computer science and machine learning and has supervised and mentored many student thesis projects at both undergraduate and graduate levels.

He is an active contributor to the research community, serving as a permanent reviewer for several respected Elsevier journals, including Machine Learning with Applications, Expert Systems with Applications and Applied Soft Computing, and as a reviewer for many international conferences. He has also served on program committees for the SST 2024 and CECNet 2025 conferences.

Alongside his academic work, Mario is an external research and development engineer at Random Red Ltd., where he combines practical software development with advanced machine learning solutions. At his Faculty, he serves as President of the Lifelong Learning Committee. He is a member of the AMA FERIT association, formerly of IEEE, and holds several professional certifications, including Microsoft Data Science and Microsoft Azure Data Scientist Associate. Mario speaks Croatian, English and Slovak.