Authors: Minrui Zheng*, University of North Carolina - Charlotte, Wenwu Tang, University of North Carolina - Charlotte
Topics: Geographic Information Science and Systems, Regional Geography, Quantitative Methods
Keywords: Multilevel model, artificial neural network, cyberinfrastructure, high-performance and parallel computing
Session Type: Paper
Start / End Time: 12:40 PM / 2:20 PM
Room: Napoleon D1, Sheraton 3rd Floor
Presentation File: No File Uploaded
Early studies were concerned with use of one statistical model to analyze spatial data. However, researchers noticed that when they attempt to create an appropriate research design, each record of a dataset is not independent existence. The entire dataset can be organized at more than one groups. These records are influenced by some groups and those groups are in turn influenced them. In other words, we need to separate the dataset base on their similarity or characteristics to attain the best performance and reduce biases. This grouping process is called multilevel analysis. There are a bunch of approaches to do multilevel analysis, for example, k-means clustering, expectation maximization clustering, or hierarchical agglomerative clustering. However, spatial data has its own characteristics, such as spatial dependence. In this study, we use multilevel analysis to examine a neural network-driven spatial modeling with the spatial error model. However, the process of analyzing big spatial data has a heavy demand of computation. Therefore, we accelerate this process using high-performance and parallel computing. Our results demonstrate that the multilevel analysis overcomes the high dimensional issue, and improves the model performance. High-performance and parallel computing approach substantially improve the computing performance.