Machine Learning and Metadata analysis to predict weevil success at controlling Eurasian Watermilfoil (EWM)

Team members: Diana White and Jon Martin (Clarkson Math), Michael Twiss (Clarkson Biology) and Thibaud Antoniou (Clarkson Data Science Major).

Studies from the past few decades suggest a correlation between the habitat of the milfoil weevil (a beetle
that specializes in feeding on all varieties of watermilfoil) and a decline in the invasive Eurasian
Watermilfoil (EWM). Due to these findings, a company called Enviro-science spent 5 years growing and
selling weevils to different lake communities, in an effort to use weevils as a bio-control for EWM. The
results of this company are mixed (approximately 50 % success rate).

In summer 2019, our team began conducting a metadata analysis on lakes for which weevils were added
by Enviro-science. From a total of 6 studies we have looked at so far, there are 78 cases where
success/failure has been recorded (all reports are from lakes throughout the US and southern Canada).
The metadata analysis comprises of a record of lake characteristics (water depth, nutrients, etc.) and the
weevil augmentation strategy (the number of weevils added, the date they were added, etc).


Using the data collected from this metadata analysis, we created a predictive classification model using
machine learning tools in Phython (scikit-learn) to relate our model predictors (lake characteristics and
augmentation strategy) to the success or failure at controlling EWM

MachineLearning_edited.jpg
MachineLearning_edited.jpg

Weevils have been found to be associated with natural decline of EWM. Augmentation efforts have shown to be inconclusive. We've completed a preliminary metadata analysis of model predictors which define success or failure in those augmentation studies.

press to zoom
ML2_edited.jpg
ML2_edited.jpg

Here, we show three examples of 2D linear predictions.

press to zoom