Combining Global Optimization Algorithms with a Simple Adaptive Distance for Feature Selection and Weighting

This work focuses on a study about hybrid optimization techniques for improving feature selection and weighting applications. For this purpose, two global optimization methods were used: Tabu Search(TS) and Simulated Annealing (SA). These methods were combined to k-Nearest Neighbor (k-NN) composing two hybrid approaches: SA/k-NN and TS/k-NN. Those approaches try to use the main advantage from the global optimization methods: they work efficiently in searching for solutions in the global space. In this study, the methodology is proposed by [4]. In the referred work, a hybrid TS/k-NN approach was suggested and successfully applied for feature selection and weighting problems. Based on the later, this analysis indicates a new SA/k-NN combination and compares their results using the classical Euclidean Distance and a Simple Adaptive Distance [8]. The results demonstrate that feature sets optimized by the studied models are very efficient when compared to the well-known k-NN. Both accuracy classification and number of features in the resultant set are considered in the conclusions. Furthermore, the combined use of the Simple Adaptive Distance improves even more the results for all datasets analyzed.