Optimization of Multilayer Perceptron (MLP) network training algorithms for agrwood oil quality separation / Noratikah Zawani Mahabob ... [et al.]

As a part of on-going research in classifying the agarwood oil quality, this research presented the optimization of the Multilayer Perceptron (MLP) network with the three different training data network algorithms; Scaled-Conjugate Gradient (SCG), Levenberg Marquardt (LM), and Resilient-Backpropagation (RBP). The work was done by using MATLAB version 2017a. The training algorithms were applied to agarwood oil data to classify its compounds to the different quality either in high or low. The data collection consists of 96 inputs of the abundances (%) of agarwood oil compounds and the output was the quality of the oil (high=2 and low=1). The process involved in data pre- processing; data normalization, data randomization, and data division. The data is divided into three groups with a ratio of 70%: 15%: 15% for training, validation, and testing respectively. The performance criteria were taken as a consideration which includes confusion matrix, accuracy, sensitivity, specificity and precision also mean square error (MSE). It was found that Levenberg-Marquardt (LM) presented the highest accuracy which was 100% for all training, validation and testing dataset with the lowest MSE. This research is important and contributed as additional research findings especially in the classification of agarwood oil area.