Tuning Prediction Model Performance using Neural Network Algorithm Learning Parameters

Accuracy rate, but with the appropriate training and learning, ANN can give a high degree of accuracy. The study had aimed to improve ANN, specifically Multilayer Perceptron, by designing a software tool that can identify a range of learning parameters' values that improves the precision of a predictive model. The study followed an experimental procedure using machine learning tools namely WEKA. Out of 15 datasets, the study had been able to identify the ranges of learning parameter values for a derived prediction model that had yielded the highest frequency with the highest precision from the identified ranges of values before the experiment. The resulting ranges under each learning parameter had been 2 to 13 for the learning rate, 61 to 84 for the momentum, 30 to 41 for the epoch, and 49 to 60 neurons for a hidden layer. Furthermore, multiple linear regression via T-Test had shown that the learning rate, momentum, and epochs did not have a significant influence on the value of the precision. The results had implied that there might be other factors that could influence the value of the precision, such as the number of instances, number of attributes, number of class, and other learning parameters such as bias.