Plant virus identification based on neural networks with evolutionary preprocessing

In this work, genetic algorithms and multilayer neural networks are applied to plant virus identification. The initial data set is derived via a well known prototype method, which uses specially designed biosensors to monitor the virus reactions. Several techniques have been introduced for preprocessing the plant virus waves. They include segmentation along the time axis for fast response, nonlinear normalization to emphasize significant information, averaging samples of the plant virus waves to suppress noise effects, reduction in the number of samples to realize a more compact network, etc. Given the features of the acquired virus time-series signals of the problem under study, an evolutionary method is proposed in order to produce meta-data from the original time-series initial information, reduce the dimensionality of the input data space, and to eliminate the noise inherent in the initial raw information. A genetic algorithm is employed so as to smooth out the initial information while, the so produced meta-data sets are used in the training and testing of the applied neural network, producing fitter training data. The method is tested against some of the most commonly used classifiers in machine learning via cross-validation and proved its potential towards assisting virus identification.