Predicting Presence of Amphibian Species Using Feature Selection

The presence of amphibian species can be regarded as the basis of any natural assessment. Generally speaking, the more amphibian species and populations analyzed, the higher the value of the habitat. In the problem of predicting presence of amphibian species, it is very difficult to assess because a large number of habitats are distributed on a vast land and the time available for field investigation is limited. The usual method is to use the local environment variable space that can be collected remotely from the satellite images and GIS systems, and combine it with machine learning method for classification and prediction. The dataset obtained in the experiments can be regarded as an ordinal classification, and some features are ordinal, which has monotonic dependence with the decision. In this paper, we introduce a feature selection algorithm to evaluate the feature space, and select sensitive features. Furthermore, we apply a machine learning algorithm to evaluate the performance of the selected feature subset, and obtain a prediction model. The experimental results show that the proposed method can effectively remove irrelevant features and improve the performance of prediction model.