The forest, along with the many products it provides, is an important source of income to the local population of the Congo Basin. Specifically, legal logging generates revenue for the local government and creates jobs for residents. However, the increase of human intrusion on the forest threatens the livelihood of ecosystems. Studies have found that the roads built by logging companies to transport logs have facilitated poaching activity in the area. Adequate land planning can be a solution to this issue. We mostly focus our work on the two tasks. First, we focus on evauating the capacity and limits of state-of-art machine learning models on predicting poaching risk using geological features. Second, given historical data, we aim at designing future land use assignments and patrol routes that would induce the least amount of poaching risk in the area. In our work, we make the following contributions: 1) we train and test several models on the task of predicting poaching risk in the Congo with multiple data sets. Our results show that with different synthetically labeled data sets, the models' performance can achieve around 0.74 in AUC. 2) we propose a suitability based forest zoning optimization problem that can assign an area with multiple land uses. 3) we propose a data-driven optimization problem to determine a set of logging sites and patrol routes that maximizes revenue while reducing poaching risk in the area. For both 2) and 3), our experiments show that our calculated solution can induce much less poaching risk in the area compared to current practice.

56 pages

Thesis Committee:
Fei Fang (Advisor)
George Chen

Srinivasan Seshan, Head, Computer Science Department
Martial Hebert, Dean, School of Computer Science

Return to: SCS Technical Report Collection
School of Computer Science

This page maintained by reports@cs.cmu.edu