Machine Learning for the Prediction of Insect Infestations in Hop Fields

European Corn Borer (ECB) is an insect that has been known as a hop pest, its infestations are one of the major problems for hop cultivation. In this thesis, several supervised machine learning techniques are applied on the data collected by the Slovenian Institute of Hop Research and Brewing for 18 years to predict the number of ECB, and the effectiveness of these techniques is compared and analyzed. The collected data consists of two parts: The first part is weather condition data which contains temperature, relative humidity, and precipitation from 8 pm to 6 am each day from April 30 to September 28, each year from 1999 to 2017; the second part is the number of ECBs captured at night on the corresponding date. The first part data is considered as the input object, while the second is considered as the desired output value. A thorough examination allowed us to conclude that regression methods are not the solution: a svm model with the help of manually added historical weather conditions information, the best out of twenty models, got an r2 score of 0.0448, only slightly better than a purely random guess. On the contrary, the work shows that the regression problem could be turned into a classification problem: predicting if the number of ECB captured is greater than a threshold. Moreover, statistics information of history can be added to the data set. Ten different kinds of classifiers and three different kinds of ways to add history information have been evaluated. Unfortunately, despite a significant improvement, results are still not accurate enough. The work shows a promising path towards the solution of this industrially-relevant problem, pinpointing deficiencies in the currently-available input data.