Risk Assessment Modeling Using Pattern Recognition and Data Mining

Abstract

Assessment of Risk is a core component of good risk management. Especially in the management of industrial and health’s risk, the accurate risk assessment methodology helps decision makers to spot hazards promptly and accurately. Once the risk is identified and assessed, determining the probability and severity of risk becomes easier. Keeping in view the two important aspects of society i.e. industrial and healthcare, both the aspects were covered by author in terms of risk assessment modeling. The industrial and health’s risk prediction aspect in terms of pursuing optimized models based on machine learning approach were covered in chapter-5 & 6 respectively. The research approach to generate risk prediction model for above both critical application aspects remained same i.e. optimized machine learning and data mining methodology to undertake risk assessment. In the industrial aspect, only textile industry was taken into account for calculating operational risk. For both the application, smart predictions systems have been proposed using friendly user interfaces. In the context of industrial risk, single machine learning technique was applied and gradually reached to proposed innovative model based on classifiers’ ensemble approach. However, keeping in view criticality of healthcare domain and practical research’s benefits to society, the healthcare application has been more focused and tested by proposed multi-layered ensemble model. In the assessment of health’s risk, two critical aspects were covered related to heart and cancer disease. Out of many millions of deaths, occurring globally from all sorts of causes, the Cardiovascular & Cancer diseases as estimated from historical figures, account for more than 20% of the overall mortalities. Nowadays many countries around the world are transforming their traditional health care system into information-based e-health system. Soft computational based e-Health-care is the process of using emerging information technologies & machine learning in health care for the benefit of humans. In the above background of soft computational based technologies and machine learning in healthcare for predictive applications, how huge medical data is effectively and efficiently “mined” from possibly multiple data sources to extract critical predictive information and consequently some innovative Data Mining methodology for above viii stated requirement with very high accuracy is the challenging research question to be efficiently addressed by machine learning methods. In the pursuit of above challenge, the current work is investigating a Multi-Layered ensemble Classifiers’ approach based on machine learning techniques for development of Intelligent Clinical Decision Support System (i-CDSS). The goal of this work is to introduce “Multi-layered ensemble classifiers methodology with multiple performance measures optimized enabling automated diagnostic framework named as Intelligent Clinical Decision Support System (i-CDSS) as the first line in the detection and diagnosis of cancer & critical heart disease. The intelligent CDSS can play a significant role in improving healthcare in terms of diagnostics’ error reduction & lessening of economic burden on patient. In this research, it is basically aimed to develop a Single Framework based on innovative data mining methodology for multiple disease diagnostics with comparatively very improved predictive accuracy as best suitable for the practical development of i-CDSS. The same work has been augmented by development of Intelligent Healthcare Application which can be implemented by materializing above framework on suitable software platform”. Multi-Layered Ensemble Classifier’s (MLEC) Model have utilized 10 independent classifiers at multiple layers of the model and further model was strengthened by combining six classifiers at each layer using Weighted voting ensemble approach thus overcoming the limitations of conventional CDSS’ performance. This research seeks to create both theoretical and practical oriented framework which is in particular applicable to general healthcare domain also. The research revealed new opportunities in the application of data mining methodology in the medical domain. We expect that this work will have a meaningful impact on the development of future health’s risk prediction system. Further it can be further extended to develop prescription models to generate critical disease treatments based on prescription data as proposed by classified medical expertise. The same can used by healthcare trainees in their initial training and house jobs in consultation with concerned medical specialist