Supervised neural network recognition of habitat zones of rice invertebrates

This study aimed to evaluate effectiveness and performance of several supervised neural network models and make pattern recognition on invertebrate habitat zones. Probabilistic, general regression, and linear neural networks, and discriminant analysis were used to recognize both known and unknown invertebrate habitat zones. The results showed that neural network models were better than traditional discriminant analysis in the recognition of known habitat zones. There was not distinctive variation in recognition from different neural network models. Sensitivity analysis indicated that the learning rate of the neural network would influence recognized results. An unknown invertebrate species from Lepidoptera was recognized to be soil-dweller (dryland) by both neural network models and discriminant analysis. In sensitivity analysis it was additionally recognized to be the type of plant canopy (terrestrial). Overall the species was estimated to be a soil-dweller (dryland) or live on plant canopy (terrestrial). It was concluded that neural network models can perform better than conventional statistic models in pattern recognition, but a comprehensive comparison among various models is necessary in order to achieve a high reliable recognition and prediction. Furthermore, sensitivity analysis can lead to an in-depth grasp on the mechanism in the recognition and is thus needed.