Tabular machine learning
Supervised learning with Scikit-learn and LightGBM. Random forests, gradient boosting and weighted ensembles, compared against clear baselines.
Model selection · Ensemble methods
MACHINE LEARNING · UBC STATISTICS
Ziyang Zhou周梓洋
Machine learning.
Statistical rigor.
I build machine learning workflows for forecasting and financial modeling. My strengths are feature engineering, tree-based ensembles and model validation, grounded in my Statistics studies at the University of British Columbia (UBC).
01 / FOCUS & METHODS
Better features. Stronger experiments.
I work across the modeling process: defining the prediction task, building useful features, comparing models and checking what holds up beyond the training data.
Features & forecasting
Turning time-series, market and business inputs into model features. Experience with electricity consumption forecasts and short-horizon stock-return ranking.
Feature engineering · Time-series prediction
Evaluation that holds up
Chronological splits, historical backtesting and checks on when information becomes available. I use validation to judge whether added complexity helps.
Temporal validation · Leakage awareness
CORE TOOLKIT
02 / FEATURED PROJECT
From prediction to portfolio decisions.
MACHINE LEARNING · FINANCIAL DATA
CSI 300 portfolio modeling
A LightGBM project for the 2026 China Collegiate Computing Contest Big Data Challenge: predict short-horizon returns, rank CSI 300 candidates and produce a five-stock allocation.
I defined the prediction task, reviewed feature availability and compared experiments across historical periods using ranking quality, returns and drawdown.
THE APPROACH
- 01Build the inputs
Align price, volume, volatility and market features with available information. - 02Challenge the model
Compare expanding training windows, with a gap before validation. - 03Make the allocation
Compare fixed and learned weights; retain the recorded LightGBM submission.
AI-assisted implementation; experiment selection and review by Ziyang Zhou.
03 / EXPERIENCE
Modeling in real business settings.
MAY & AUG 2026
Data Analyst Intern
Anhui Shusheng Data Technology
安徽数升数据科技有限公司
Contributed to weekly and monthly electricity forecasts: screened candidate features, helped build and tune random forests, and compared weighted ensembles through historical backtesting.
JUN – JUL 2026
Actuarial Department Intern
Shanghai Guangbo Hi-intelligence Technology
上海光博高智能科技有限公司
Supported GLM-based insurance pricing with SAS: helped validate risk factors, compared model specifications, checked rate-table factors and co-authored a standard operating manual.
JUL – AUG 2025
IPO Project Intern
Guotai Haitong Securities
国泰海通证券
Supported a Beijing Stock Exchange IPO project through industry research, financial and comparable-company analysis, and prospectus data verification.
04 / 中文简介
周梓洋
Ziyang Zhou · UBC 统计学
以统计学为基础，
专注机器学习与预测建模。
我是周梓洋（Ziyang Zhou），英属哥伦比亚大学（UBC）统计学本科生，预计于 2027 年 5 月毕业。我专注于应用机器学习，擅长特征工程、树模型与集成学习，以及模型验证；使用 Python、Scikit-learn 和 LightGBM，将预测问题转化为可比较、可检验的建模流程。
我的实践涵盖用电量预测、沪深 300 股票收益排序与投资组合建模。我重视数据的时间边界、基线比较和历史回测，通过实验判断新增特征或复杂模型是否真正有效。
我曾在安徽数升数据科技有限公司参与随机森林与加权集成预测，在上海光博高智能科技有限公司精算部参与 GLM 保险定价支持，并在国泰海通证券参与 IPO 项目实习。
05 / EXPLORE & CONNECT
The models. The experiments.
The work behind them.
Project case studies, source code and professional experience.