I'm a passionate data scientist with strong training in statistics and a proven track record in data competitions. My recent focus has been on time series forecasting, stock return prediction, and motion sensor data classification. I enjoy tackling real-world problems with machine learning, statistical modeling, and collaborative teamwork.

我是一位對資料科學充滿熱情的統計背景研究生，擁有多場數據競賽的實作經驗。研究興趣涵蓋時間序列預測、股價報酬建模，以及運動感測資料的分類問題，並樂於在團隊中整合模型設計與報告溝通。

👉 Download My Resume (PDF)

• M.S. in Statistics and Data Science, National Tsing Hua University (2023–2025)
• B.S. in Applied Mathematics, National Dong Hwa University (2018–2022)

• 🔍 Stock-Price-Prediction with Industry Partner (2024)

  Collaborated with UMC (United Microelectronics Corporation) in a team-based stock price forecasting project. Contributed to the design and implementation of the LSTM-based prediction model, combined with log-returns and GARCH-estimated volatility to construct 95% confidence intervals. Participated in technical discussions with the company and took responsibility for summarizing and presenting part of the final project report. Achieved 80% directional accuracy in next-day stock movement forecasting.

  參與與聯華電子（UMC）合作之股價預測團隊專案，負責設計與實作 LSTM 模型，結合 log-return 與 GARCH 模型建構 95% 預測區間。參與企業端技術討論，並負責部分成果彙整與簡報，次日股價漲跌預測準確率達 80%。

👉 相關細節可見下方的「📂 Selected Projects | 精選專案」。

• ✨ Top 5% in AI CUP 2025: Table Tennis Smart Racket Classification Built XGBoost + Voting models on time-series sensor data, incorporating FFT and ADASYN techniques. Collaborated with a cross-disciplinary team, half of whom came from non-statistical backgrounds, and led the modeling strategy and communication.

  於 AI CUP 桌球競賽中排名前 5%，主導使用 FFT、小波、小樣本增強與 XGBoost+Voting 架構達成穩定成效，並與多元背景隊友合作（其中約一半非統計科學背景），展現良好的團隊協作與溝通能力。

• 🌟 Top 3% in SinoPac AI GO 2025: Soaring Stock Prediction Challenge
  From over 10,000 variables, selected 23 key features by combining multiple feature importance evaluation methods and taking their intersection. Applied ADASYN to address class imbalance, and integrated models using a stacking framework. Designed a one-pass SOP that effectively tackled two major challenges in the competition.

  於 2025 永豐銀行 AI GO 飆股預測競賽排名前 3%，結合多種變數重要性評估方式並取其交集，從上萬個變數中精選出 23 個關鍵特徵，使用 ADASYN 處理類別不平衡，並以 stacking 架構整合多模型。進一步提出一套 SOP，有效且一次性地解決競賽中所面對的兩大挑戰。

👉 相關細節可見下方的「📂 Selected Projects | 精選專案」。

• Languages: Python, SQL, R
• Courses Taken: Deep Learning, Machine Learning, Time Series Analysis, Applied Multivariate Analysis, Categorical Data Analysis
• ML Techniques: Time Series Forecasting (LSTM, AR models), Tree-based Models (XGBoost, Random Forest), Model Selection (Lasso, Forward/Backward Selection, XGBoost Importance, Random Forest Importance, Permutation Importance), Model Ensembling (Stacking, Voting), Imbalanced Learning (ADASYN, SMOTE)

• 📈 Stock Price Prediction (Industry Collaboration)
  Applied LSTM to predict whether the stock price would rise or fall the next day, and constructed a 95% confidence interval using GARCH-estimated volatility. Achieved 80% directional accuracy.
  📌 中文摘要：使用 LSTM 預測明日股價漲跌方向，並結合 GARCH 模型估計波動度、建構 95% 預測區間，漲跌預測準確率達 80%。

• 📊 Soaring Stock Prediction Challenge (SinoPac AI GO 2025)
  Developed a feature selection and stacking framework to handle a high-dimensional stock dataset. From over 10,000 variables, selected 23 key features—approximately one-tenth the number used by typical competitors—via multiple feature importance evaluation techniques. The final model ranked in the top 3% of the public leaderboard.
  📌 中文摘要：參與「2025 永豐 AI GO 飆股預測競賽」，結合多種特徵重要性評估方式，從上萬個變數中精選出 23 個關鍵特徵（約為其他參賽者的十分之一），並透過 stacking 架構整合模型，最終於公開組排名前 3%。

• 🏓 Table Tennis Smart Racket Classification (AI CUP 2025)
  Performed feature extraction using FFT & wavelets, handled class imbalance with ADASYN, and applied XGBoost with ensemble voting. Final result ranked in the top 5% of the public leaderboard.
  📌 中文摘要：針對 AI CUP 桌球資料，整合小波與傅立葉特徵，搭配 ADASYN 處理不平衡問題，以集成模型完成分類，最終成績於公開組排名前 5%。

• 🎮 LOL How to Win?
  Developed a predictive model for League of Legends tournament outcomes. Achieved 73% accuracy, outperforming baseline by 23%, and identified that only two key variables are sufficient for quick win-loss prediction.
  📌 中文摘要：開發《英雄聯盟》比賽預測模型，準確率達 73%，相較基準提升 23%，並歸納出僅需兩個關鍵變數組合即可快速預測勝負。

• Email: a10155105105@gmail.com
• GitHub: github.com/ILaskira