Discretizing Whale Optimization Algorithm to Optimize a Long Short-Term Memory

If previous data can influence data, then the data can be said to have sequential properties. Unlike the data that is not sequential, the randomization of sequential data sequences can change the data. A common neural network model generally cannot distinguish a sequential data from the non-sequential ones. Thus, the recurrent model is made specifically for managing data with sequential properties that must be considered by studying the relationship between data with the previous ones. To create a recurrent model, some parameters should be carefully designed. One of them is the architecture of the model. In this paper, discretizing the whale optimization algorithm (WOA), which is performed by determining the hidden layer number of neurons and dropout an architecture, is proposed to optimize the long short-term memory (LSTM). Evaluations on the large movie review dataset show that the proposed discrete WOA is capable of significantly giving an absolute improvement of the LSTM mean accuracy by up to 1.50% (from 91.23% to 92.73%).