همایش بین المللی هوش مصنوعی

صفحه اصلی / دومین همایش بین المللی هوش مصنوعی

Hybrid Machine Learning and Deep Learning Approach for Minimum Airfare Prediction

نویسندگان :

Mohammad Safaiyan¹ Ali Mahvash Mohammadi²

1- Isfahan University of Technology 2- university of Isfahan

کلمات کلیدی :

Minimum airfare prediction،Hybrid ensemble learning،XGBoost،LSTM،1D-CNN،Dynamic pricing systems،Regression Weighted Accuracy (RWA)،Large-scale time series

چکیده :

Airfare prediction is a challenging task due to dynamic pricing mechanisms, limited real‑time visibility into fare updates, and nonlinear demand–supply interactions. This paper focuses on Minimum Fare Prediction: forecasting the lowest airfare a route will reach within the next three days so that users can decide whether to buy now or wait. We propose a hybrid framework that integrates gradient‑boosted tree ensembles (XGBoost and Histogram Gradient Boosting Regressor) with temporal neural architectures (1D‑CNN, LSTM, and a CNN→LSTM hybrid) trained on a unified feature space. The models are trained and evaluated on a proprietary Pateh.com dataset of approximately 49.7 million flight search records collected over 480 days, one of the largest corpora analyzed in airfare prediction research. Each instance corresponds to the minimum observed daily fare for a specific origin–destination route, enriched with 20 engineered features including calendar descriptors, lagged fares, and rolling 3‑day minimum prices. To evaluate practical usefulness, we introduce Regression Weighted Accuracy (RWA), a tolerance‑based metric that rewards predictions within a ±10% band of the true price and penalizes larger deviations. Using a time‑series cross‑validation scheme with 12 folds, our hybrid CNN→LSTM model achieves higher RWA than existing baselines such as k‑nearest neighbors, gradient boosting, and bagging regression trees on a key route (IFN–DXB), and reaches up to 0.60 RWA on selected high‑traffic routes. The approach is currently being A/B tested within Pateh’s production system to provide “buy now vs. wait” recommendations and display the predicted minimum fare for the next three days.

لیست مقالات

لیست مقالات بایگانی شده

From RLHF to DPO and Beyond: A Review of Preference Optimization Methods for Aligning Large Language Models

Poorya Saneei - Parham Rahimi - Behrouz Minaei-Bidgoli

PerSHOP: A Multi-Domain Persian Dataset for Shopping Dialogue Systems

Keyvan Mahmoudi - Heshaam Faili

AI-Driven Materials Genome: Accelerated Discovery of Multi-Functional Sensing Materials

Farzane Hasheminia - Sadegh Sadeghzadeh - Behrouz Minaei-Bidgoli

Calibrating Large Language Models for High-Quality Automated Writing Feedback

Mohammad Sajjad Aghadadi - Heshaam Faili

White blood cell image analysis using CNN model

Fahimeh Jahanbakhshi - Hamid Latifi

An Extended Table Detection Framework for Enhanced Recognition of Persian Tables in OCR Systems

Aras Valizadeh - Bita Baroutian - Fateme Taraghi - Zahra Mohammadi - Atefe Aghaei - Mohsen Ebrahimi Moghaddam

Examining Ethical Principles in the Development of AI for Environmental Protection with a Focus on Environmental Justice

Maryam Saadaat Nabavi Meybodi

Fibroglandular Tissue Classification in Breast MRI: A Comparative Study of Automated Decision Strategies

Meysam Khalaj - Arvin Arian - Ala Torabi - Nasrin Ahmadinejad - Masoumeh Gity - Seyedeh Nooshin Miratashi Yazdi - Mohammad Pooya Afshari - Melika Sadeghi Tabrizi - Hamid Soltanian-Zadeh

MQL-NPC: A Modified Q-Learning-based Approach to Design Intelligent Non-Player Character in a Survival Game

Morteza Nalbandi - Athena Abdi

Enhancing IoT Data Prediction Accuracy Using Deep Learning and Metaheuristic Algorithms

Safoura Ashoori - Khadigh Nemati - Mohamad hadi Amini

بیشتر