Pengembangan Metaheuristik Protis pada Neural Network untuk Mengoptimalkan Lapisan Tersembunyi (Hidden Layer)

Abstract

The difficulty of determining the hidden layer neurons in classical neural network architecture is a frequent problem, so an in-depth exploration and exploitation approach is needed with the Protis Metaheuristic and Swarm Intelligence method approaches by adapting the amoeba protist lifestyle where there are 4 phases, namely prophase, metaphase, anaphase, and telophase, where the 4 phases are modified in the neural network architecture. To optimize the right number of hidden layers and produce a good neural network architectural model with optimal Performance. The research aims to develop a mathematical formulation into the process of searching for neurons in the hidden layer according to the metaheuristic approach, Swarm Intelligence, protist theory to optimize the hidden layer, and the best architecture in the process of training and testing the neural network model, then to compare the Performance of the classification results using the protis and other classical neural network methods in terms of Model Training process, interpretation, Performance and stability, among others, Feed Forward Neural Networks, Recurrent Neural Networks and Convolutional Neural Networks, especially for time series, categorical and image data as well as Comparing the protis neural network method with metaheuristic methods, other swarm intelligence. The dataset used is Categorical, Time Series, and image. The research results show that the Protis Neural Network can be implemented on Time Series, Categorical, and Image Data types. The protest method can optimize forming neuron architecture in the hidden layer. The average range of the number of neurons in the hidden layer for a neural network is 0 – 35 neurons in the layer. The accuracy level of the Protis algorithm embedded in the Neural Network for Categorical and time series data, the average Precision value achieved is 0.952, and Recall reaches 0.950. The accuracy level of the Protis algorithm embedded in the Protis Convolutional Neural Network average value reaches 95.9%, so from the three data tested, the Protis Convotional gets the best accuracy value.