Analyzing Prediction Performance between Wavelet Neural Network and Product-Unit Neural Network

Abstract

Analyzing the performance of a particular approach in a field very dependent on the problem it's aimed to solve. Artificial Neural Network (ANN) widely used for prediction in many areas including medical, environment, business intelligence and education. The uniqueness of ANN is the dynamic of hidden layer can be improvised mapped with the data problem and the structure of architecture can be enhanced such as Wavelet Artificial Neural Network (WANN) and Product Unit Neural Network (PUNN). This research aimed to analyzed the performance between WANN and PUNN towards water quality data of Chini Lake. Real world data comes with dynamic stream data and dynamic parameters based on its area of data collection method. Handling dynamic data would be misleading if the approach used very dependent towards data classes. The measurement to analyze the data based on performance accuracy, data sensitivity, data precision and specification of both method with regards of the regular ANN. The findings demonstrate the ability to obtain satisfactory prediction accuracy for both WANN and PUNN compared to regular ANN. The model accuracy for this case study by using WANN and PUNN were 75.34 % and 66.86 %, respectively. Therefore, WANN would be a competitive tool for prediction with conventional ANN.