Airfare Price Prediction Based on TPA-Transformer

doi:10.11871/jfdc.issn.2096-742X.2023.06.011

Abstract

Abstract:

[Objective] The airline industry is one of the industries that use the most complex pricing strategies. Even the ticket prices on the same flight fluctuate dynamically and significantly. Passengers can only make decisions based on experience, such as buying tickets as soon as possible, but it is not always the best choice. The traditional price forecasting methods cannot sufficiently capture the dependence between complex internal/external factors and the ticket prices. [Methods] This paper designs and implements the TPA-Transformer (Ticket Price Aware Transformer) to predict ticket price and proposes a related data processing method based on time series. This model adds the attention module to introduce reference information and the multi-layer convolution after the encoder to fuse the information of different features of multiple flights and extract local features to improve the model's performance in multi-step price prediction. [Results] The model is verified on five regression evaluation indexes (MSE, RMSE, MAE, ACC, and AMS). [Conclusions] Experiments show that the model effectively improves the prediction accuracy and is superior to other five comparison models (Random Forest, XGBoost, LSTM, GRU, and Transformer).

Key words: airfare price prediction, machine learning, time series, attention mechanism

SHEN Zhihao, LI Na, YIN Shihao, DU Yi, HU Lianglin. Airfare Price Prediction Based on TPA-Transformer[J]. Frontiers of Data and Computing, 2023, 5(6): 115-125, https://cstr.cn/32002.14.jfdc.CN10-1649/TP.2023.06.011.

Figures/Tables 11

Fig.1

Fig.2

Fig.3

Fig.4

Table 1

Fig.5

Table 2

Table 3

One step prediction experimental result"

数据集	模型	MSE $↓$	ACC (%) $↑$	RMSE $↓$	MAE $↓$	AMS $↑$
单航班	Random Forest	19731	89.41	140.46	118.26	58.69
	XGBoost	19973	90.05	141.32	104.53	62.61
	LSTM	16769	91.74	129.49	100.46	65.38
	GRU	16890	92.02	129.96	96.09	64.47
	Transformer	10347	92.57	101.72	82.95	67.05
	TPA-Transformer	8779	93.66	93.69	66.41	68.62
多航班	Random Forest	22642	88.56	150.47	131.59	55.71
	XGBoost	18737	89.86	136.88	117.28	62.69
	LSTM	15046	90.63	122.66	106.72	60.69
	GRU	14286	90.93	119.18	105.29	61.68
	Transformer	13195	91.94	114.86	97.08	65.46
	TPA-Transformer	10253	92.76	101.25	78.35	66.46

Table 3

Fig.6

Fig.7

Fig.8

References 16

[1]	ABDELLA J A, ZAKI N M, SHUAIB K, et al. Airline ticket price and demand prediction: A survey[J]. Journal of King Saud University-Computer and Information Sciences, 2021, 33(4): 375-391. doi: 10.1016/j.jksuci.2019.02.001
[2]	李晓花, 萧柏春. 航空公司收入管理价格与舱位控制的统一分析[J]. 管理科学学报, 2004, 7(6): 63-69.
[3]	席卫东, 乔兵, 朱剑英, 等. 引入乘客博弈的民航收益管理决策优化[C]// 中国优选法统筹法与经济数学研究会第七届全国会员代表大会暨第七届中国管理科学学术年会论文集, 2005: 223-227.
[4]	HOPMAN D, KOOLE G, MEI R V D. A machine learning approach to itinerary-level booking prediction in competitive airline markets[J]. International Journal of Revenue Management, 2021, 12(3-4): 153-191. doi: 10.1504/IJRM.2021.120347
[5]	ETZIONI O, TUCHINDA R, KNOBLOCK C A, et al. To buy or not to buy: mining airfare data to minimize ticket purchase price[C]// Proceedings of the ninth ACM SIGKDD international conference on Knowledge discovery and data mining, 2003: 119-128.
[6]	GROVES W, GINI M. On optimizing airline ticket purchase timing[J]. ACM Transactions on Intelligent Systems and Technology (TIST), 2015, 7(1): 1-28.
[7]	LANTSEVA A, MUKHINA K, NIKISHOVA A, et al. Data-driven modeling of airlines pricing[J]. Procedia Computer Science, 2015, 66(1): 267-276. doi: 10.1016/j.procs.2015.11.032
[8]	LI Y, LI Z. Design and implementation of ticket price forecasting system[C]// AIP Conference Proceedings, AIP Publishing LLC, 2018, 1967(1): 040009.
[9]	LIU T, CAO J, TAN Y, et al. ACER: An adaptive context-aware ensemble regression model for airfare price prediction[C]// 2017 International Conference on Progress in Informatics and Computing (PIC), IEEE, 2017: 312-317.
[10]	TZIRIDIS K, KALAMPOKAS T, PAPAKOSTAS G A, et al. Airfare prices prediction using machine learning techniques[C]// 2017 25th European Signal Processing Conference (EUSIPCO), IEEE, 2017: 1036-1039.
[11]	SUBRAMANIAN R R, MURALI M S, DEEPAK B, et al. Airline Fare Prediction Using Machine Learning Algorithms[C]// 2022 4th International Conference on Smart Systems and Inventive Technology (ICSSIT), IEEE, 2022: 877-884.
[12]	LU J. Machine learning modeling for time series problem: Predicting flight ticket prices[J]. arXiv preprint arXiv:1705.07205, 2017.
[13]	DU K, YAN J, HANG Z, et al. An lstm based deep learning method for airline ticket price prediction[C]// International Conference on Neural Information Processing, Springer, Cham, 2020: 762-769.
[14]	张前前. 基于机器学习的机票购买时机预测[D]. 北京: 北京邮电大学, 2020.
[15]	ZHAO Z, YOU J, GAN G, et al. Civil airline fare prediction with a multi-attribute dual-stage attention mechanism[J]. Applied Intelligence, 2021, 1(1): 1-16.
[16]	WEN Q, ZHOU T, ZHANG C, et al. Transformers in Time Series: A Survey[J]. arXiv preprint arXiv:2202.07125, 2022.

属性	描述	范围
机票价格	所选时间节点的真实价格	价格浮点数
起飞日期	航班预定起飞日期	1~361
距离起飞天数	购票时距离起飞剩余天数	0~20
航班号	航班所属航班号，One-Hot	0,1
是否节假日	起飞当天是否为节假日，One-Hot	0,1
周几	起飞日期在每周第几天，One-Hot	0,1
起飞时间	起飞当日具体时间，0~24.00	浮点数
降落时间	降落当日具体时间0~24.00	浮点数

参数	描述	值
learning rate	学习率	1e-4
d_model	输入维度	48
Batchsize	批数据大小	64
n_head	多头注意力头数	4
num_encoder_layers	Encoder层数	6
Dropout	丢弃率	0.1
Optimizer	优化器	Adam
n_head(attention module)	参考信息多头注意力头数	4