Finding the Critical Feature Dimension of Big Datasets

Abstract

Big Data allied to the Internet of Things nowadays provides a powerful resource that various organizations are increasingly exploiting for applications ranging from decision support, predictive and prescriptive analytics, to knowledge and intelligence discovery. In analytics and data mining processes, it is usually desirable to have as much data as possible, though it is often more important that the data is of high quality thereby raising two of the most important problems when handling large datasets: sample and feature selection. This work addresses the sampling problem and presents a heuristic method to find the “critical sampling” of big datasets. The concept of the critical sampling size of a dataset is defined as the minimum number of examples that are required for a given data analytic task to achieve a satisfactory performance. The problem is very important in data mining, since the size of data sets directly relates to the cost of executing the data mining task. Since the problem of determining the optimal solution for the Critical Sampling Size problem is intractable, in this dissertation a heuristic method is tested, in order to infer its capability to find practical solutions. Results have shown an apparent Critical Sampling Size for all the tested datasets, which is rather smaller than the their original sizes. Further, the proposed heuristic method shows a promising utility, providing a practical solution to find a useful critical sample for data mining tasks.

Big Data allied to the Internet of Things nowadays provides a powerful resource that various organizations are increasingly exploiting for applications ranging from decision support, predictive and prescriptive analytics, to knowledge and intelligence discovery. In analytics and data mining processes, it is usually desirable to have as much data as possible, though it is often more important that the data is of high quality thereby raising two of the most important problems when handling large datasets: sample and feature selection. This work addresses the sampling problem and presents a heuristic method to find the “critical sampling” of big datasets. The concept of the critical sampling size of a dataset is defined as the minimum number of examples that are required for a given data analytic task to achieve a satisfactory performance. The problem is very important in data mining, since the size of data sets directly relates to the cost of executing the data mining task. Since the problem of determining the optimal solution for the Critical Sampling Size problem is intractable, in this dissertation a heuristic method is tested, in order to infer its capability to find practical solutions. Results have shown an apparent Critical Sampling Size for all the tested datasets, which is rather smaller than the their original sizes. Further, the proposed heuristic method shows a promising utility, providing a practical solution to find a useful critical sample for data mining tasks.