A classification of sequential patterns for numerical and time series multiple source data — A preliminary application on extreme weather prediction

Classification based on sequential patterns has become very important method in data mining. It is useful to make predictions for alert warning system and strategic decision. Moreover the necessity to improve the speed performance of sequential pattern mining also increases. However, previous researches on this area uses categorical data as input. There is necessity to process numerical data and classify sequential patterns found from the data. High accuracy numerical data are difficult to mine. Moreover, numerical data to be mined consist of many observational parameter data. This study proposes framework to overcome the problem. The framework proposes to categorize the data in preprocessing and prepare it to be ready as input for sequential pattern mining and the subsequent classification process. The framework will improve classification speed, scalability and also maintain the classification accuracy.