High-dimensionality and class imbalance represent two main challenges in classification. Because of the curse of dimensionality, the performance of many classification algorithms is often decreased. When datasets are unbalanced, some classification algorithms might suffer from a performance bias, often achieving a relatively poor minority accuracy, while in many real world applications, the minority accuracy is often at least as important as the majority accuracy. Recently, a growing number of high-dimensional data sets are unbalanced, resulting in more challenging classification tasks since these datasets may exhibit the characteristics of a combination of the class imbalance and high-dimensionality. Genetic programming (GP) is often effective as a classification method and also as a promising approach to feature construction. However, most existing GP methods are also biased toward the majority class once the data distribution is unbalanced. This research focuses on GP for classification with high-dimensional unbalanced data. The overall goal is to develop a new GP approach to effectively and efficiently enhancing both the minority and majority accuracies,and to overcoming the curse of dimensionality issue. This will be achieved by investigating new fitness functions for GP, developing new multi-objective GP methods for classification and feature construction, and developing cost-sensitive GP methods