Abstract:
This research aims to build a model for paper usage estimation within airlines passenger
service department using Artificial Neural Network(Multi-layer Perceptron) which consist of 42
attributes and 256 data records. The neural network multi-layers perceptrons will learn 4 data sets
first the raw data, second the filtered data from regression analysis method, third the filtered data
with 10 attributes derived from Information Gain method and the last one with 10 attributes
derived from Gain Ratio method by the multi-layer perceptron with backpropagation learning.
Then compare each model to find the least value of Mean Square Error (MSE) will be brought to
test to find the least Mean Absolute Percentage Error (MAPE) by comparing the target value with
the estimated value from neural network. The results showed that suitable model for further
development is the Regression Analysis attributes selection method have attribute are training
scale, number of operation days, number of staff training, porter, important arrival-departure
information, departure information and year. The 3-layer neural networks, the Tansig-Logsig
active function, and Pureline output layer 30-15-1 provide the lowest value is 4.3 of MAPE,
0.000854 of MSE value. Because small network structure or number of node use less variables
and fast calculate performance.