Active Learning to Maximize Area Under the ROC Curve

In active learning, a machine learning algorithmis given an unlabeled set of examples U, and is allowed to request labels for a relatively small subset of U to use for training. The goal is then to judiciously choose which examples in U to have labeled in order to optimize some performance criterion, e.g. classification accuracy. We study how active learning affects AUC. We examine two existing algorithms from the literature and present our own active learning algorithms designed to maximize the AUC of the hypothesis. One of our algorithms was consistently the top performer, and Closest Sampling from the literature often came in second behind it. When good posterior probability estimates were available, our heuristics were by far the best.

Distribution of Rain Gardens in Lincoln Nebraska: Are Rain Gardens more Likely to be Built Near Bodies of Water

Abstract Rain gardens are an important tool in reducing the amount of stormwater runoff and accompanying pollutants from entering the city’s streams and lakes, and reducing their water quality. This thesis project analyzed the number of rain gardens installed through the City of Lincoln Nebraska Watershed Management’s Rain Garden Water Quality Project in distance intervals of one-eighth mile from streams and lakes. This data shows the distribution of these rain gardens in relation to streams and lakes and attempts to determine if proximity to streams and lakes is a factor in homeowners installing rain gardens. ArcGIS was used to create a map with layers to determine the number of houses with rain gardens in 1/8 mile distance increments from the city’s streams and lakes and their distances from a stream or lake. The total area, number of house parcels, and the type and location of each parcel type were also determined for comparison between the distance interval increments. The study revealed that fifty-eight percent of rain gardens were installed within a quarter mile of a stream or lake (an area covering 60% of the city and including 58.5% of the city’s house parcels), and that eighty percent of rain gardens were installed within three-eighth mile of streams or lakes (an area covering 75% of the city and 78.5% of the city’s house parcels). All parcels in the city are within 1 mile of a stream or lake. Alone the number of project houses per distance intervals suggested that proximity to a stream or lake was a factor in people’s decisions to install rain gardens. However, when compared to the number of house parcels available, proximity disappears as a factor in project participation.