What We Learned from the 2025 China University Big Data Competition

Team Overview

Team Name: BDC大王来了

Members: Zhuang Minglei (Hebei Agricultural University), Gu Jinhang (Hebei Agricultural University), Wang Jingyi (Hebei Agricultural University)

National Ranking: Top 5

Competition Brief

The 2025 China University Computer Competition – Big Data Challenge invites teams to tackle a financial forecasting problem. Participants must predict not only price movements but also the top‑10 and low‑10 ranked stocks, making ranking accuracy a crucial metric.

Key Technical Takeaways

Data Pitfalls & Feature Engineering: Financial benchmark data often contain hidden traps; effective feature engineering, especially extracting trend indicators, can reveal market dynamics more clearly than raw prices.

Align Model Design with Scoring: The scoring emphasizes ranking (top‑10/low‑10) rather than simple price change, so models should be optimized for ranking accuracy, potentially yielding better results.

Generalization Over Precision: Final scores are evaluated on a hidden test set, so over‑fitting to the development data is risky. Training multiple models on data subsets and averaging predictions can improve robustness.

Leverage SOTA Models Rationally: Review state‑of‑the‑art solutions from similar stock‑prediction contests, adopt proven architectures, and then fine‑tune them for the specific competition requirements.

Competition Experience

The organizers provided the best support the team has ever experienced, offering comfortable accommodation, abundant meals, and direct interaction with Tsinghua University professors during meals, which helped alleviate stress.

Teams from across the country displayed diverse strengths—some focused on graduate research, others on innovative projects or early‑year enthusiasm—creating a friendly, collaborative atmosphere both onstage and offstage.

The team expressed gratitude to the organizers, co‑organizers, and judges for the valuable opportunity and guidance, wishing continued success for future editions of the competition.

Images