Classification and prediction of social attributes By K-Nearest Neighbor Algorithm with Socially-aware wireless networking-A study

Abstract

Abstract In this paper, we analyze the data which was collected from SIGCOMM 2009, where 76 users were taken part in and traces of Bluetooth encounters, opportunistic messaging, and social profiles of them were collected. The experiment started by conducting a the data sorting and data cleaning. In data processing, mainly three main types of characteristic data in the dataset: message, proximity and participant were chosen. After that, Hierarchical cluster analysis was applied on the processed data, which can score the selected segments by similarity measurement, and then form and visually describe the hierarchical structure of these selected clusters. T can either be manually set or a machine language algorithm can be used and here the manually number of categories are set to be 3 after applying the fviz_nbclust function for optimal clustering analysis. As this was a study article five different calculation methods were analyzed to accomplish hierarchical clustering: single linkage, complete linkage, median distance method, UPGMA and Centroid Clustering. The comparison is deployed iwith a simulation tool and the results are plotted. After the clustering algorithms, the participants are divided into three categories, and for each clustering algorithm the divisions are different. In this regard, k-nearest neighbor classification algorithm is used to calculate the error rate of comparison between the original data and the predicted data. To conclude, Our method results in specific label classification and then analyze the strength of nodes and the tightness between nodes via the KNN algorithm.