Geographically Weighted Discriminant Analysis (GWDA) is a method for prediction and analysis of categorical spatial data. It is an extension of Linear Discriminant Analysis (LDA) that allows the relationship between the predictor variables and the categories to vary spatially. This is also referred to spatial non-stationarity. If spatial non-stationarity exists, GWDA should model the relationship between the categories and predictor variables more accurately, thus resulting in a lower classification uncertainty and ultimately a higher classification accuracy. The GWDA output also requires interpretation to understand which variables are important in driving the classification in different geographical regions. This research uses interactive visualisations from the field of geovisual analytics to investigate the performance of GWDA in terms of classification accuracy, classification uncertainty and spatial non-stationarity. The methodology is demonstrated in a case study that uses GWDA to examine the relationship between county level voting patterns in the 2004 US presidential election and five socio-economic indicators. This research builds on existing techniques to interpret the GWDA output and provides additional insight into the processes driving the classification. It also demonstrates a practical application of geovisual analytic tools.