Question: If an evolutionary algorithm is a general-purpose search and optimization technique, it can be applied to determine suitable network architectures and weight values for a neural network. Implement an evolutionary algorithm to determine a suitable MLP network architecture trained with the backpropagation algorithm to solve the IRIS problem described below. There are a vast number of data sets available for testing and benchmarking neural networks and learning algorithms. For instance, the University of California Irvine (UCI) provides many collections of data sets accessible via anonymous FTP at ftp.ics.uci.edu or their website at https://www.ics.uci.edu/~mlearn/MLRepository.html.
The IRIS data set is a very well-known classification problem containing 150 samples divided into three differences classes, each of which with 50 patterns. The samples have a dimension of 4 (m = 4) and refer to a type of iris of plants. One of the three classes is linearly separable from the others, and two of them are not. Further documentation about this problem can be downloaded from the UCI repository together with the data set. Tip: define a suitable representation for the chromosomes, a fitness function, and appropriate genetic operators. Each chromosome will correspond to a neural network that will have to be trained via backpropagation in order to be evaluated. The objective is to find a network with a classification error less than 5%.