Imbalanced Learning with Parametric Linear Programming Support Vector Machine For Weather Data Application

Abstract

Learning from imbalanced data sets is one of the aspects of predictive modeling and machine learning that has taken a lot of attention in the last decade. Multiple research projects have been carried out to adjust the existing algorithms for accurate predictions of both classes. The model proposed in this thesis is a linear Support Vector Machine model with L1-norm objective function with applications on weather data collected from the Bureau of Meteorology system in Australia. Apart from model selection and modifications we have also introduced a parametric modeling algorithm based on a novel parametric simplex approach for parameter tuning of Support Vector Machine. The combination of the two proposed approaches has yielded a significant improvement in predicting the minority class and decrease the model’s bias towards the majority class as is seen in most machine learning algorithms.