The prospective use of machine learning in combatting wildlife trafficking in Cameroon

Abstract

Wildlife populations have reached a critical point in Central Africa with many species facing points of no return. Often, this can be attributed to the illicit wildlife trade. Many states in the region are currently facing elevated levels of corruption and violence, resulting in a lack of adequate anti-trafficking efforts from authorities. As such, ambitious initiatives and ideas to combat this illicit trade are needed before key species become extirpated. Focusing on data sciences, because of rapid advancements in this field and its increasing accessibility, the potential for its application to the Illicit Wildlife Trade (IWT) is being more frequently explored. Machine learning is a prime example of a field that has been revolutionized over the last decade in conjunction with other technological tools such as mapping, data visualization, and data analysis. Machine learning has reached a point where, when combined with these other tools, has the potential to assist in combating this trade through examining its past, present, and future. While there are many instances where machine learning has been utilized, including in analyzing IWT, the application of this tool to wildlife trafficking at a country level was not identified as having been done during the research phase of this project. In this research, we discovered that current publicly available datasets proved vastly inadequate for use in this study. This factor led to the creation of a custom dataset of trafficking instances in Cameroon. The data was then analyzed utilizing tools such as ARC GIS Online and Microsoft PowerBI. Based off this initial analysis, both non-machine learning and machine learning techniques were employed for further analysis and in specific cases to forecast. What resulted was models that struggled due to the limitations of the dataset, with the models affected by overfitting. What can be concluded from this study is that while not achieved here, it is possible to forecast (to some degree) areas prone to trafficking. The lack of data means that these forecasts can only be so useful and their reliability for actual policy formulation can be called into question. Perhaps more valuable are the trends identified from the more basic analysis, as these allow for more concrete conclusions to be reached. The final conclusion that can be reached is that, without an adequately sized dataset, the application of machine learning, while proven, cannot alone be utilized to combat the illicit wildlife trade in Cameroon or Central Africa.