Scalability of Data Mining Algorithms for Non-Stationary Data

Streaming data is increasing day-by-day in real world applications such as stock market, financial data etc. The prime quality of this data is the online advent and collection at high speed and vulnerability to transforms in the data distributions due to the dynamic environment. The biggest challenge of analyzing this data is the lack of unrestricted availability of non-stationary dataset for evaluation and comparison. Extracting useful information from non-stationary dataset is also challenging because of its scaling nature. In this paper, different classification techniques of data mining are applied for analyzing non-stationary dataset using different existing and proposed cascade scaling method. The approach begins with the collection of a real time dataset of National Stock Exchange of India NSEI from online Yahoo finance site, which is then pre-pro cessed. The novel approach of three rules has been applied to fix the binary target value based on the buy/sell of the shares on the particular day. Once the dataset has been pre-processed, it is decomposed into smaller datasets of equal size subsets. The particular data mining approach is then applied identically to each subset. The findings of the data mining approach on all subsets are pooled and aggregated for the final output. The performance of the suggested algorithm is assessed using a number of criteria, including accuracy, precision, recall, f-score, and execution time. The comparative analysis of the mentioned performance metrices are evaluated for the various data mining approaches. The next comparative analysis includes the non-scaling and proposed cascade scaling method for the various data mining approaches. Lastly, the performance comparison of existing scaling methods with proposed cascade scaling method for the accuracy parameter is evaluated where proposed method achieved substantial performance on the non-stationary dataset.