Sašo Moškon (2009) Nomogram-based search for subspaces of independent attributes. EngD thesis.
In thesis we introduce selective nomograms, an improvement of nomograms for visualization of naive Bayesian classifier. Selective nomograms allow us to interactively explore the domain and discover conditional dependencies between the attributes. We also propose a classification algorithm based on the idea of selectable nomograms. First, we introduce selective nomograms, define conditional dependencies and describe the theoretical background for discovering conditional dependencies between the attributes using selective nomograms. We present experiments and empirically evaluate selective nomograms. Then we propose the idea of using selective nomograms for searching the neighborhood of given example. We present an implementation of searching the neighborhood and describe how it can be used as a classification method. We empirically evaluate the classifier and discuss the results. Finally we propose some ideas for future work.
Actions (login required)