Ventilator control with ieRSPOP++ and ieRSPOP (FRI/E) using interpolative and extrapolative reasoning

• Main Author: Tun, Zin Zin
• Other Authors: Quek Hiok Chai
• Format: Final Year Project
• Language: English
• Published: 2016
• Subjects: DRNTU::Engineering
• Online Access: http://hdl.handle.net/10356/66949
• Institution: Nanyang Technological University

Artificial Intelligent is the rising technique that have been widely used in expert task domain, like trading analysis/decision and medical sector. Many different learning algorithm have emerged over past few decades. Fuzzy logic is known for its high interpretability. Therefore, Fuzzy logic & Neural Network are combined to have an understandable reasoning and conclusion. Pseudo outer-Product Fuzzy Neural Network (POPFNN) with Rough Set attribute reduction (RSPOP) have shown to have achieved high interpretability with moderately high accuracy. The improved online learning ieRSPOP (incremental ensemble learning RSPOP) handle well time series data and the result are promising. ieRSPOP is further improved into two ways, ieRSPOP++ and ieRSPOP with interpolation and extrapolation (ieRSPOP FRI/E). The former method is to improve accuracy and interpretability with anti-hebbian learning and Ebbinghaus theory for dynamic updating of ensemble weights. The latter is used interpolation and extrapolation method to handle sparse rule based condition. ieRSPOP FRI/E have shown that its achievement over time series data like Mackey Glass time series and stock market prediction to other methods. However, the application of it to medical sector is still lacking. In this paper, author investigated these two architectural functionalities, providing the actual implemental architectural flow of ieRSPOP FRI/E, it will server a better understanding for the existing ieRSPOP family structure and deep insight of the method explained. The ieRSPOP FRI/E experiment on artificial ventilation modelling is done and strength and weakness of existing system is identified. Recent researches have shown that people are interested in handling sparse data because these data can make an impactful decision for financial and medical expert task. Therefore, how these two architectural can be further improvement when dealing with sparse data is discussed for future development as well.