Estimation of Australia’s Outbound Airline Passenger Demand Using an Adaptive Neuro-Fuzzy Inference System

Abstract

This study has proposed and empirically tested an adaptive neuro-fuzzy inference system (ANFIS) model for predicting Australia's outbound international airline passenger demand. The model was developed using eleven input parameters of world GDP, world population, world air fare yields, world jet fuel prices, outbound flights from Australia, Australia's unemployment numbers, Australian's (AUD/USD) foreign exchange rate, Australia's outbound tourist expenditure and four dummy variables. The model was constructed using annual data from 1994 to 2019. The hybrid learning algorithm and the subtractive clustering partition method were used to generate the optimum ANFIS models. The performance of the model was measured using five error measures: coefficient of determination (R2-value), root mean square errors (RMSE), mean absolute errors (MAE) and the mean absolute percentage error (MAPE). The results found that the mean absolute percentage error (MAPE) for the overall data set of the model was 3.60%. The R2-value was around 0.9886, demonstrating that the ANFIS is an efficient model for predicting Australia's outbound airline passenger demand.

Predicting Australia’s Domestic Airline Passenger Demand using an Anfis Approach

“…Following training, the ANFIS model for forecasting Australia's domestic enplaned passengers was validated by selecting 16 data points, which are different from the other 93 points used for adaptive neuro fuzzy inference system (ANFIS) training (Srisaeng and Baxter, 2021;Srisaeng et al, 2015b). Each validation data point was fed into the system and then Australia's predicted domestic airline enplaned passengers was computed and compared to the actual values.…”

“…Prior to training the data in the adaptive neuro-fuzzy inference system (ANFIS), it is important to process the data into patterns. The normalization of the data ensures that the ANFIS will be trained effectively and will prevent any variable skewing the results significantly (Srisaeng et al, 2015b;Srisaeng and Baxter, 2021). Consequently, all the input parameters are of equal importance in training the artificial neural network (ANN) system within the adaptive neuro fuzzy inference system (ANFIS) (Srisaeng and Baxter, 2021).…”

“…Data normalization was applied to transform the data into a symmetric distribution (Grosche et al, 2007). This practice improves model performance since the data appear to satisfy the assumptions of a statistical inference procedure more closely (Srisaeng et al, 2015b;Srisaeng and Baxter, 2021). The data collected for the study were normalized using the following equation:…”

“…In this process an array of training errors was obtained for evaluation (Chen et al, 2010). In this study, the testing data subset was kept separate to the training data set and was used to train the adaptive neuro fuzzy inference system (ANFIS) model (Srisaeng and Baxter, 2021). The data was therefore divided into two randomly selected groups: the first group of 77 data was used as the training set (85% of the overall data), and the remaining 16 data was used for verifying and testing the robustness of the adaptive neuro fuzzy inference system ANFIS) based prediction model (Savkovic et al, 2019, Srisaeng et al, 2015b.…”

“…The data was therefore divided into two randomly selected groups: the first group of 77 data was used as the training set (85% of the overall data), and the remaining 16 data was used for verifying and testing the robustness of the adaptive neuro fuzzy inference system ANFIS) based prediction model (Savkovic et al, 2019, Srisaeng et al, 2015b. The adaptive neuro fuzzy inference system (ANFIS) model used 77 training data in 1-400 training epochs (Srisaeng and Baxter, 2021). Figure 3 shows the training curve of the adaptive neuro fuzzy inference system (ANFIS) model which has a root mean square error (RMSE) of 0.001162.…”