Saved in:
Bibliographic Details
Main Authors: Liu, Fei, Ren, Huanhuan, Guan, Yu, Wang, Xiuxu, Lv, Wang, Hu, Zhiqiang, Chen, Yaxi
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2505.07245
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909607601897472
author Liu, Fei
Ren, Huanhuan
Guan, Yu
Wang, Xiuxu
Lv, Wang
Hu, Zhiqiang
Chen, Yaxi
author_facet Liu, Fei
Ren, Huanhuan
Guan, Yu
Wang, Xiuxu
Lv, Wang
Hu, Zhiqiang
Chen, Yaxi
contents Predicting future vehicle purchases among existing owners presents a critical challenge due to extreme class imbalance (<0.5% positive rate) and complex behavioral patterns. We propose REMEDI (Relative feature Enhanced Meta-learning with Distillation for Imbalanced prediction), a novel multi-stage framework addressing these challenges. REMEDI first trains diverse base models to capture complementary aspects of user behavior. Second, inspired by comparative op-timization techniques, we introduce relative performance meta-features (deviation from ensemble mean, rank among peers) for effective model fusion through a hybrid-expert architecture. Third, we distill the ensemble's knowledge into a single efficient model via supervised fine-tuning with MSE loss, enabling practical deployment. Evaluated on approximately 800,000 vehicle owners, REMEDI significantly outperforms baseline approaches, achieving the business target of identifying ~50% of actual buyers within the top 60,000 recommendations at ~10% precision. The distilled model preserves the ensemble's predictive power while maintaining deployment efficiency, demonstrating REMEDI's effectiveness for imbalanced prediction in industry settings.
format Preprint
id arxiv_https___arxiv_org_abs_2505_07245
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle REMEDI: Relative Feature Enhanced Meta-Learning with Distillation for Imbalanced Prediction
Liu, Fei
Ren, Huanhuan
Guan, Yu
Wang, Xiuxu
Lv, Wang
Hu, Zhiqiang
Chen, Yaxi
Machine Learning
Artificial Intelligence
Predicting future vehicle purchases among existing owners presents a critical challenge due to extreme class imbalance (<0.5% positive rate) and complex behavioral patterns. We propose REMEDI (Relative feature Enhanced Meta-learning with Distillation for Imbalanced prediction), a novel multi-stage framework addressing these challenges. REMEDI first trains diverse base models to capture complementary aspects of user behavior. Second, inspired by comparative op-timization techniques, we introduce relative performance meta-features (deviation from ensemble mean, rank among peers) for effective model fusion through a hybrid-expert architecture. Third, we distill the ensemble's knowledge into a single efficient model via supervised fine-tuning with MSE loss, enabling practical deployment. Evaluated on approximately 800,000 vehicle owners, REMEDI significantly outperforms baseline approaches, achieving the business target of identifying ~50% of actual buyers within the top 60,000 recommendations at ~10% precision. The distilled model preserves the ensemble's predictive power while maintaining deployment efficiency, demonstrating REMEDI's effectiveness for imbalanced prediction in industry settings.
title REMEDI: Relative Feature Enhanced Meta-Learning with Distillation for Imbalanced Prediction
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2505.07245