Metaheuristics for Mining Massive Datasets: A Comprehensive Study of PSO for Classification

摘要

The increasing importance of Meta - Heuristic Algorithms for Data Mining Problems ha s paved way for the emergence of a new era of computing algorithms that performs the functionalities of legacy algorithms using meta heuristic techniques. The occurrence of this change was due to the massive increase in the data that is used for analysis a nd the inability of the legacy techniques to either incorporate huge data, or the increase in the time consumed by these algorithms. This paper deals with probabilistic nature of PSO and to understand the error and how it helps in obtaining efficient solut ions. A discussion on why PSO was selected as the base algorithm for our study, and its nuances are discussed in detail. The paper also deals with techniques to overcome the local optima and the scalability and parallelization options available in PSO . The discussion ends with metric wise comparisons of PSO with a variety of datasets and how PSO for classification provides better stability and accuracy on massive datasets rather than smaller ones justifying the study.