Machine Learning-based Event Data Mining in Healthcare and Manufacturing
Event data, encompassing real-world occurrences with specific topics, locations, and timeframes, plays a pivotal role in diverse fields such as healthcare, manufacturing, and business, etc. Efficient and accurate event data mining is vital for deriving valuable insights and improving decision-making processes. However, event data mining faces unique challenges that require the development and integration of advanced machine learning techniques. These challenges include the risk of privacy leakage in multi-source event data mining, the difficulties in capturing various characteristics of events, and the complex dependencies among multichannel multitype events generated from multivariate time series. This dissertation addresses these challenges and contributes to event data analytics in the healthcare and manufacturing domains. First, we propose novel multi-source event data mining approaches that protect sensitive information while retaining analytical utility. Second, we develop advanced machine learning techniques for heterogeneous multi-output prediction of events, enabling accurate prediction of various characteristics simultaneously. Third, we introduce innovative methodologies for discovering complex dependencies among multichannel multitype events, providing insights into their intricate relationships and underlying mechanisms in multivariate time series. By integrating and developing state-of-the-art machine learning-based techniques, this research advances the field of event data mining, offering significant benefits to the healthcare and manufacturing sectors by enhancing decision-making processes and overall performance.