This document discusses feature engineering and decision tree modeling on a forest cover type dataset. It describes the dataset attributes, various preprocessing experiments including soil type and wilderness area generalization, and new engineered features. Decision trees C4.5, C5.0 and CART were evaluated, with C5.0 achieving the highest accuracy of 91.11%. Further tuning C5.0 with pruning and using ensemble methods like random forest and boosted C5.0 improved accuracy to 77.24% and 76.02% respectively.