A Genetic Algorithm Approach for Semi-Supervised Clustering Algorithm

Abstract

A semi-supervised clustering algorithm is proposed that combines the benefits of supervised and unsupervised learning methods. The approach allows unlabeled data with no known class to be used to improve classification accuracy [2]. The objective function of an unsupervised technique, e.g. K-means clustering, is modified to minimize both the cluster dispersion of the input attributes and a measure of cluster impurity based on the class labels. Minimizing the cluster dispersion of the examples is a form of capacity control to prevent over fitting [4]. For the output labels, impurity measures from decision tree algorithms such as the Gini index can be used. A genetic algorithm optimizes the objective function to produce clusters. Experimental results show that using class information improves the generalization ability compared to unsupervised methods based only on the input attributes [6]. Training using information from unlabeled data can improve classification accuracy on that data as well. Genetic Algorithms (GAs) have been widely used in optimization problems for their high ability in seeking better and acceptable solutions within limited time. Clustering ensemble has emerged as another flavour of optimal solutions for generating more stable and robust partition from existing clusters [1]. GAs has proved a major contribution to find consensus cluster partitions during clustering ensemble. Currently, web video categorization has been an ever challenging research area with the popularity of the social web. In this paper, we propose a framework for web video categorization using their textual features, video relations and web support [3]. There are three contributions in this research work. First, we expand the traditional Vector Space Model (VSM) in a more generic manner as Semantic VSM (S-VSM) by including the semantic similarity between the features terms [5]. This new model has improved the clustering quality in terms of compactness (high intracluster similarity) and clearness (low inter-cluster similarity). Second, we optimize the clustering ensemble process with the help of GA using a novel approach of the fitness function. We define a new measure, PrePaired Percentage (PPP), to be used as the fitness function during the genetic cycle for optimization of clustering ensemble process [7]. Third, the most important and crucial step of the GA is to define the genetic operators, crossover and mutation. We express these operators by an intelligent mechanism of clustering ensemble. This approach has produced more logical offspring solutions [9]. Above stated all three contributions have shown remarkable results in their corresponding areas. Experiments on real world social-web data have been performed to validate our new incremental novelties [8].