Design and development of neural Bayesian approach for unpredictable stock exchange databases

Predicting stock market has been a well researched topic in the field of financial engineering. However, most methods suffer from serious drawback due to handling uncertain and missing continuous time-series data. We divide our study in two modules: data cleaning and decision making, first module includes preprocessing of stock market time series data mining which presents the extended form of extraction of both graphical structure and conditional probabilities of a Bayesian Belief Networks (BBN) from a possible incomplete stock market time series databases. In second decision making module, we introduce Linguistic Rules-Tree (LR-Tree) which is the combination of fuzzy logics and decision tree but LR-Tree may not be the best generalization due to over-fitting. Consequently Neuro-Pruning method has been introduced for post pruning of LR-Tree. In Neuro-Pruning, instead of absolutely removing nodes, we employ a back-propagation neural network to give weights to nodes according to their significance. After data cleaning and decision making modules we hope that, our proposed approaches will be able to handle attributes with differing costs, improving computational efficiency, outperforms error-based pruning, handle uncertain and missing continuous time-series data.