Hybrid Prediction of the Appearance of Monkeys using Markov Chain Model and Support Vector Machine

Abstract

Recently, agricultural damage caused by harmful animals, especially monkeys, is a critical problem in Japan. This paper proposes a system which predicts dates that monkeys approach farmland. In order to make predictions, monkeys’ activities were collected around a mountain for two years. As a result of the investigation, the monkeys were appeared on several points, and it was expected that they were moving according to specific pattern. Therefore, in this study, the Markov chain model that can handle state transitions stochastically was adopted as the method for monkeys’ behavior prediction. As a result of calculating the Markov chain on the order from 1st to 5th, 2nd order was optimal for this study. It could be considered that the monkeys were moving a specific pattern at a cycle of a few days. In the case of a two-class problem, i.e. monkeys appear or not, 57.5 % accuracy was obtained. In the multi-class problem, which place monkeys appear, the accuracy was 31.5 %. Additionally, the hybrid of Markov chain and Support Vector Machine brings more efficiency of the monkeys’ appearance prediction.