🧠 Daily Study Log [2025-07-05]
Today’s focus was SCU_Competition experimentation (submission 56–60), plus structured TOEIC study (Listening Unit 2, Reading Unit 2–3).

📊 SCU_Competition — Submission 56–60

Focus: StackingClassifier revisited, cluster augmentation, noise filtering, and explosive feature engineering
Goal: Push leaderboard AUC to new peak via structure vs feature trade-off

Highlights:

• Submission 56: StackingClassifier with LGBM meta-model
  → CV AUC: 0.8617 / Kaggle AUC: 0.8898
• Submission 57: Voting (5:3:2) with added 구매패턴_클러스터
  → CV AUC: 0.8807 / Kaggle AUC: 0.8956
• Submission 58: Voting with cluster count increased (n_clusters=6)
  → CV AUC: 0.8808 / Kaggle AUC: 0.8948
• Submission 59: Pruned bottom 10% of features by importance
  → CV AUC: 0.8816 / Kaggle AUC: 0.8953
• Submission 60: Added 10+ new derived features (ratios, interactions)
  → CV AUC: 0.8788 / Kaggle AUC: 0.8973 🔥

Next Ideas:

• Try simplified model using top-30 features
• OneHot encode cluster labels
• Test LGBM+RF-only ensemble (drop LR)

📘 TOEIC Study — Listening & Reading Practice

Focus Areas:

• 🎧 Listening Unit 2: Short Conversations
  • Practiced identifying speaker intent, setting, and logical flow
  • Noted traps around similar-sounding words and indirect suggestions
• 📖 Reading Unit 2–3: Vocabulary-in-Context & Inference
  • Focused on identifying transition words (e.g., however, therefore)
  • Applied skimming/scanning to long paragraphs for targeted answers
  • Improved confidence on inference-type questions

Reflection:

• Listening speed still fast but improving
• Reading was much smoother today — vocabulary context clues are getting clearer
• Unit-based practice feels more productive than random full-tests

✅ TL;DR

📍 SCU: Submission 60 nearly beat the all-time best — feature engineering wins
📍 TOEIC: Focused on Listening Unit 2 + Reading Unit 2–3 — better structure & flow