Experimental design for feature matching of geospatial data: results of the quality control service
This dataset contains the results of the experimental design for feature matching presented in the thesis 'Automatic evaluation of geospatial data quality using web services'. Design of experiment techniques were developed to assess how some variables are influenced by a list of factors. In this study we chose four variables: precision, recall, F-measure, and time. We selected six factors that might influence these variables: similarity measures, matching methods, morphology of features, geographic context, systematic disturbance, and random disturbance. Regarding our four variables, precision and recall are concepts that come from the Information Retrieval field. Precision evaluates the presence of wrong matches (false positives) against the real matches (true positives). Recall evaluates the presence of non-matches (false negatives) against those real matches. F-measure represents the harmonic mean between precision and recall. Time represents the time consumed to run some matching procedure. In a previous study we identified that the geospatial data matching problem can be organized in two key aspects: similarity measures and matching methods. So these are the two first factors to be investigate: measures and methods. The third controlled factor is the morphology of objects, i.e., how the roughness of lines or areas influences the variables. Other factor is the geographic context of features, which refers to the spatial relationships between objects in a neighbourhood. The last two factors refers to some disturbances applied to source data: systematic and random perturbations. This experiment is divided according to the geometric primitive: point, line, and area. Figure Fig_Overview_of_DOE.png shows an overview of this DOE with the factors considered for each geometry, and the respective number of treatments for each essay. Each type of geometry has its own essays, 20 in total: points (P1-P5), lines (L1-L6), and areas (A1-A9). The results of each essay are presented in CSV format: blank space as separator, first line as header.
Steps to reproduce
The results of each essay can be load into the R statistical computing tool (https://www.r-project.org/) using the read.table method. Example: in order to load the data from the essay P1 the user should use: df <- read.table("/path/to/data/P1.csv", header=TRUE) This command loads all data to the data frame named 'df'. The variables are always named 'precision', 'recall', and 'time'. Not all experiments contains 'time'. The other columns represent the factors that influence the variables, which can vary from essay to essay. F-measure can be calculated from precision and recall using: df$fmeasure <- ifelse( df$precision > 0, 2*df$precision*df$recall/(df$precision+df$recall), 0 )