\n | 파일 | 주요 컬럼 | 집계 방향 |
|---|
\n \n | customers | \n age, loyalty_score, email_open_rate,
discount_usage_rate, avg_review_score, customer_tier, credit_limit | \n as-is (1:1) — customer_id별 | \n
\n \n | transactions | \n order_value, items_count, payment_method,
discount_applied, shipping_speed, high_value_flag | \n 고객별 집계 | \n
\n \n | sessions | \n session_duration, pages_viewed, cart_additions,
bounce_flag, traffic_source, device_type | \n 고객별 집계 | \n
\n \n | geo_data | \n average_income, urban_ratio,
internet_penetration, region_tier | \n geo_ip_region 기준 join | \n
\n \n | marketing | \n campaign_type, campaign_budget | \n campaign_id 기준 → sessions에 join 후 집계 | \n
\n
\n\n\n\n\n**[데이터 타입&결측치]**\n\n| no | 항목 | 수치 | 처리 방향 |\n|---|---|---|---|\n| ① | 데이터 shape | (200,000 × 16) | — |\n| ② | email_open_rate dtype | float64 | — |\n| ③ | email_open_rate 결측 | 45,823건 / 22.91% | 동일 고객 그룹 + 최근 구매일 + 동일 review 그룹 기준 대체 예정 |\n| ④ | discount_usage_rate 결측 | 38,756건 / 19.38% | 처리 방법 검토 중 |\n| ⑤ | 두 컬럼 동시 결측 | 12,345건 / 6.17% | — |\n\n\n\n\n## 3. 접근법 ##\n\n#### ML 문제 유형 → 이진 분류 (Binary Classification)\n\n\n\n**[Top 5 예측 변수]**\n\n| 순위 | 변수 | 선택 근거 |\n|---|---|---|\n| ① | loyalty_score | cltv가 높을 확률. 가장 직관적이고 강력한 예측력 기대 |\n| ② | recency_days | \"얼마나 최근에 구매했는가\"가 30일 내 구매 예측과 가장 연관성 높을 것으로 예상 |\n| ③ | customer_tier | 4개 카테고리 — 등급별 구매 패턴 분석 예정 |\n| ④ | avg_review_score | 만족도가 재구매 의향에 직결. 낮은 등급 고객은 재구매 의사 낮을 것으로 판단 |\n| ⑤ | shipping_speed | 배송 속도 민감 고객 파악. 만족도 높을수록 플랫폼 이탈 가능성 낮음 |\n\n\n\n**[변수 분류]**\n\n**① RFM 기반 피처** (transactions에서 생성)\n- Recency : 마지막 거래일 ~ 기준일까지 일수\n- Frequency : 고객별 총 거래 횟수\n- Monetary : 총 주문 금액 합계 → 구매 예측 모델에서 가장 강력한 예측력 예상\n\n**② 고객 행동 지표** (customers)\n- loyalty_score, email_open_rate, discount_usage_rate, avg_review_score\n- customer_tier (Bronze/Silver/Gold/Platinum) → ordinal encoding 적용\n\n**③ 지역 특성** (geo_data + customers.region join)\n- average_income, urban_ratio, internet_penetration\n- region_tier (Tier1~3) → 지역별 구매력·접근성 반영\n\n**④ 거래 세부 행동** (transactions)\n- payment_method 다양성 (crypto 사용 여부 등)\n- shipping_speed, high_value_flag, discount_applied 비율, items_count\n\n\n**[noise 컬럼 검토]**\n\n- 상관관계 아직 명확히 파악되지 않음. cardinality 높음 (약 150개 ID)\n- ID별 대략적인 그래프를 그려 추세 확인 → noise 그래프와 각 변수 그래프 간 유사한 흐름 존재하나 완전히 동일하지 않음\n- 일부 구간에서 영향 없는 것으로 확인 → **추가 검증 후 제거 여부 결정 예정**\n\n\n\n## 4. 기대 결과 & 리스크 ##\n\n**[주요 리스크]**\n\n- 클래스 불균형 | 현실 이커머스 데이터는 가입자 대비 활성 사용자가 5% 미만 가능\n → 대부분의 target이 0 | class_weight / SMOTE / scale_pos_weight 적용 \n- 데이터 품질 - 결측치 다수 (email_open_rate 23%, discount_usage_rate 19%)\n →커스터머 그룹 기반(평균) 대체 or missing flag 피처 생성 \n- 과적합 - 피처 수 51개, 트리 계열 모델 → Cross-validation + Early stopping\n\n\n**[6주 실행 계획]**\n\n| 주차 | 목표 |\n|---|---|\n| 1주 | 데이터 탐색 & 전처리 |\n| 2주 | 피처 엔지니어링 1 |\n| 3주 | 피처 엔지니어링 2 |\n| 4주 | 베이스라인 모델 (LightGBM) &모델 튜닝 & 피처 선택 |\n| 5주 | 앙상블 & 검증 |\n| 6주 | 발표 자료 & 최종 제출 |\n","metadata":{}},{"cell_type":"code","source":"import os\nprint(\"Input 폴더 확인\")\nprint(os.listdir('/kaggle/input'))\n\nprint(\"\\n대회 데이터 폴더 확인\")\nfor folder in os.listdir('/kaggle/input'):\n if 'e-commerce' in folder.lower() or 'purchase' in folder.lower():\n print(\"✅ 찾음:\", folder)\n path = f'/kaggle/input/{folder}'\n print(\"파일 목록:\", os.listdir(path))\n break","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2026-04-04T08:26:43.251849Z","iopub.execute_input":"2026-04-04T08:26:43.252779Z","iopub.status.idle":"2026-04-04T08:26:43.260060Z","shell.execute_reply.started":"2026-04-04T08:26:43.252744Z","shell.execute_reply":"2026-04-04T08:26:43.258629Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"import pandas as pd\nimport os\n\n# competitions 안에 있는 대회 폴더 찾기\ncomp_path = '/kaggle/input/competitions'\n\nprint(\"Competitions 폴더 확인\")\nprint(os.listdir(comp_path))\n\n# 대회 폴더 이름 찾기\nfor folder in os.listdir(comp_path):\n if 'e-commerce' in folder.lower() or 'purchase' in folder.lower():\n data_path = f'{comp_path}/{folder}/'\n print(\"✅ 대회 폴더 찾음:\", folder)\n print(\"파일 목록:\", os.listdir(data_path))\n \n # train과 test 불러오기\n train = pd.read_csv(data_path + 'train.csv')\n test = pd.read_csv(data_path + 'test.csv')\n \n print(\"\\nTrain shape:\", train.shape)\n print(\"Test shape :\", test.shape)\n print(\"Train 컬럼:\", train.columns.tolist())\n break\n ","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2026-04-04T08:27:45.460918Z","iopub.execute_input":"2026-04-04T08:27:45.461319Z","iopub.status.idle":"2026-04-04T08:27:45.550019Z","shell.execute_reply.started":"2026-04-04T08:27:45.461289Z","shell.execute_reply":"2026-04-04T08:27:45.548851Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"import pandas as pd\n# 전처리 1차 \ndata_path = '/kaggle/input/competitions/global-e-commerce-customer-purchase-prediction/'\n\n# customers.csv 불러오기\ncustomers = pd.read_csv(data_path + 'customers.csv')\n\nprint(\"customers shape:\", customers.shape)\nprint(\"customers 컬럼:\", customers.columns.tolist())\nprint(\"\\n처음 5행:\")\nprint(customers.head())\n\n# train과 customers 합치기 (customer_id 기준)\ntrain = pd.read_csv(data_path + 'train.csv')\nmerged = train.merge(customers, on='customer_id', how='left')\n\nprint(\"\\nmerged shape:\", merged.shape)\nprint(merged.head())\n","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2026-04-04T08:29:46.473530Z","iopub.execute_input":"2026-04-04T08:29:46.474557Z","iopub.status.idle":"2026-04-04T08:29:47.911338Z","shell.execute_reply.started":"2026-04-04T08:29:46.474506Z","shell.execute_reply":"2026-04-04T08:29:47.910267Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"import pandas as pd\n# 전처리 1차 후 간단한feature 탐색\n\ndata_path = '/kaggle/input/competitions/global-e-commerce-customer-purchase-prediction/'\n\n# 파일 불러오기\ntrain = pd.read_csv(data_path + 'train.csv')\ncustomers = pd.read_csv(data_path + 'customers.csv')\ntransactions = pd.read_csv(data_path + 'transactions.csv')\nsessions = pd.read_csv(data_path + 'sessions.csv')\n\nprint(\"불러오기 완료\")\nprint(\"train:\", train.shape)\nprint(\"customers:\", customers.shape)\nprint(\"transactions:\", transactions.shape)\nprint(\"sessions:\", sessions.shape)\n\n# 1. customers + train 합치기\ndf = train.merge(customers, on='customer_id', how='left')\n\n# 2. 간단한 전처리 (noise 컬럼 제거 + 날짜 처리)\nnoise_cols = [col for col in df.columns if 'noise_customer' in col]\ndf = df.drop(columns=noise_cols, errors='ignore')\n\n# signup_date를 날짜로 변환\ndf['signup_date'] = pd.to_datetime(df['signup_date'], errors='coerce')\n\nprint(\"\\n기본 merge 후 shape:\", df.shape)\nprint(\"컬럼:\", df.columns.tolist())\nprint(df.head())","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2026-04-04T08:30:47.703044Z","iopub.execute_input":"2026-04-04T08:30:47.703436Z","iopub.status.idle":"2026-04-04T08:30:54.619497Z","shell.execute_reply.started":"2026-04-04T08:30:47.703407Z","shell.execute_reply":"2026-04-04T08:30:54.618245Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"import pandas as pd\n\ndata_path = '/kaggle/input/competitions/global-e-commerce-customer-purchase-prediction/'\n\ntransactions = pd.read_csv(data_path + 'transactions.csv')\n\nprint(\"transactions 컬럼 목록:\")\nprint(transactions.columns.tolist())\nprint(\"\\n처음 3행:\")\nprint(transactions.head(3))","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2026-04-04T08:37:57.501539Z","iopub.execute_input":"2026-04-04T08:37:57.502569Z","iopub.status.idle":"2026-04-04T08:37:58.419364Z","shell.execute_reply.started":"2026-04-04T08:37:57.502531Z","shell.execute_reply":"2026-04-04T08:37:58.418238Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"import pandas as pd\nfrom datetime import datetime\n\ndata_path = '/kaggle/input/competitions/global-e-commerce-customer-purchase-prediction/'\n\ntrain = pd.read_csv(data_path + 'train.csv')\ncustomers = pd.read_csv(data_path + 'customers.csv')\ntransactions = pd.read_csv(data_path + 'transactions.csv')\n\n# 기본 merge\ndf = train.merge(customers, on='customer_id', how='left')\n\n# noise 컬럼 제거\nnoise_cols = [col for col in df.columns if 'noise' in col]\ndf = df.drop(columns=noise_cols, errors='ignore')\n\n# transaction_timestamp → datetime 변환\ntransactions['transaction_timestamp'] = pd.to_datetime(transactions['transaction_timestamp'], errors='coerce')\n\n# order_value 전처리: $ 기호와 쉼표(,) 제거 후 숫자로 변환\ntransactions['order_value'] = (\n transactions['order_value']\n .astype(str)\n .str.replace('$', '', regex=False) # $ 제거\n .str.replace(',', '', regex=False) # 쉼표 제거\n .astype(float)\n)\n\n# 고객별 RFM 계산\nrfm = transactions.groupby('customer_id').agg({\n 'transaction_timestamp': lambda x: (datetime.now() - x.max()).days, # Recency\n 'transaction_id': 'count', # Frequency\n 'order_value': ['sum', 'mean'] # Monetary\n}).reset_index()\n\nrfm.columns = ['customer_id', 'recency', 'frequency', 'monetary_sum', 'monetary_mean']\n\n# df에 RFM 붙이기\ndf = df.merge(rfm, on='customer_id', how='left')\n\nprint(\"RFM 추가 후 shape:\", df.shape)\nprint(\"\\nRFM 앞 5행:\")\nprint(df[['recency', 'frequency', 'monetary_sum', 'monetary_mean']].head())\nprint(\"\\nRFM 요약 통계:\")\nprint(df[['recency', 'frequency', 'monetary_sum', 'monetary_mean']].describe())","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2026-04-04T08:41:12.834215Z","iopub.execute_input":"2026-04-04T08:41:12.834626Z","iopub.status.idle":"2026-04-04T08:41:34.011368Z","shell.execute_reply.started":"2026-04-04T08:41:12.834589Z","shell.execute_reply":"2026-04-04T08:41:34.010365Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"import lightgbm as lgb\nfrom sklearn.model_selection import train_test_split\nfrom sklearn.metrics import roc_auc_score\nimport warnings\nwarnings.filterwarnings('ignore')\n\n# 전처리: age와 credit_limit 숫자로 변환\ndf['age'] = pd.to_numeric(df['age'], errors='coerce') # unknown → NaN\ndf['credit_limit'] = df['credit_limit'].astype(str).str.replace('$', '', regex=False).str.replace(',', '', regex=False)\ndf['credit_limit'] = pd.to_numeric(df['credit_limit'], errors='coerce')\n\n# 사용할 피처 선택\nfeature_cols = ['age', 'loyalty_score', 'email_open_rate', 'discount_usage_rate', \n 'avg_review_score', 'credit_limit', 'recency', 'frequency', \n 'monetary_sum', 'monetary_mean']\n\nX = df[feature_cols]\ny = df['target']\n\n# Train / Valid split\nX_train, X_valid, y_train, y_valid = train_test_split(\n X, y, test_size=0.2, random_state=42, stratify=y\n)\n\nprint(\"LightGBM 학습 시작...\")\n\nlgb_train = lgb.Dataset(X_train, y_train)\nlgb_valid = lgb.Dataset(X_valid, y_valid, reference=lgb_train)\n\nparams = {\n 'objective': 'binary',\n 'metric': 'auc',\n 'learning_rate': 0.05,\n 'num_leaves': 64,\n 'max_depth': -1,\n 'subsample': 0.8,\n 'colsample_bytree': 0.8,\n 'n_estimators': 1000,\n 'verbose': -1,\n 'random_state': 42,\n 'class_weight': 'balanced'\n}\n\nmodel = lgb.train(\n params,\n lgb_train,\n valid_sets=[lgb_valid],\n callbacks=[lgb.early_stopping(100), lgb.log_evaluation(100)]\n)\n\n# Valid 점수\nvalid_pred = model.predict(X_valid)\nprint(\"\\nValid AUC:\", roc_auc_score(y_valid, valid_pred))","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2026-04-04T08:43:15.425551Z","iopub.execute_input":"2026-04-04T08:43:15.425881Z","iopub.status.idle":"2026-04-04T08:43:17.311787Z","shell.execute_reply.started":"2026-04-04T08:43:15.425853Z","shell.execute_reply":"2026-04-04T08:43:17.310794Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"\n# (RFM + LightGBM)\n\nimport pandas as pd\nimport lightgbm as lgb\nfrom datetime import datetime\n\ndata_path = '/kaggle/input/competitions/global-e-commerce-customer-purchase-prediction/'\n\n# 데이터 불러오기\ntrain = pd.read_csv(data_path + 'train.csv')\ntest = pd.read_csv(data_path + 'test.csv')\ncustomers = pd.read_csv(data_path + 'customers.csv')\ntransactions = pd.read_csv(data_path + 'transactions.csv')\n\n# 1. customers merge\ntrain_df = train.merge(customers, on='customer_id', how='left')\ntest_df = test.merge(customers, on='customer_id', how='left')\n\n# noise 제거\nnoise_cols = [col for col in train_df.columns if 'noise' in col]\ntrain_df = train_df.drop(columns=noise_cols, errors='ignore')\ntest_df = test_df.drop(columns=noise_cols, errors='ignore')\n\n# 2. order_value 전처리 ($와 쉼표 제거)\ntransactions['order_value'] = (\n transactions['order_value']\n .astype(str)\n .str.replace('$', '', regex=False)\n .str.replace(',', '', regex=False)\n .astype(float)\n)\n\n# transaction_timestamp 변환\ntransactions['transaction_timestamp'] = pd.to_datetime(transactions['transaction_timestamp'], errors='coerce')\n\n# 3. RFM 계산 (train + test 공통)\nrfm = transactions.groupby('customer_id').agg({\n 'transaction_timestamp': lambda x: (datetime.now() - x.max()).days,\n 'transaction_id': 'count',\n 'order_value': ['sum', 'mean']\n}).reset_index()\n\nrfm.columns = ['customer_id', 'recency', 'frequency', 'monetary_sum', 'monetary_mean']\n\n# RFM merge\ntrain_df = train_df.merge(rfm, on='customer_id', how='left')\ntest_df = test_df.merge(rfm, on='customer_id', how='left')\n\n# 4. age, credit_limit 숫자로 변환\ntrain_df['age'] = pd.to_numeric(train_df['age'], errors='coerce')\ntest_df['age'] = pd.to_numeric(test_df['age'], errors='coerce')\n\ntrain_df['credit_limit'] = train_df['credit_limit'].astype(str).str.replace('$', '', regex=False).str.replace(',', '', regex=False)\ntest_df['credit_limit'] = test_df['credit_limit'].astype(str).str.replace('$', '', regex=False).str.replace(',', '', regex=False)\n\ntrain_df['credit_limit'] = pd.to_numeric(train_df['credit_limit'], errors='coerce')\ntest_df['credit_limit'] = pd.to_numeric(test_df['credit_limit'], errors='coerce')\n\n# 5. 학습용 피처\nfeature_cols = ['age', 'loyalty_score', 'email_open_rate', 'discount_usage_rate', \n 'avg_review_score', 'credit_limit', 'recency', 'frequency', \n 'monetary_sum', 'monetary_mean']\n\nX = train_df[feature_cols]\ny = train_df['target']\nX_test = test_df[feature_cols]\n\n# Train/Valid split\nX_train, X_valid, y_train, y_valid = train_test_split(X, y, test_size=0.2, random_state=42, stratify=y)\n\n# LightGBM 학습\nlgb_train = lgb.Dataset(X_train, y_train)\nlgb_valid = lgb.Dataset(X_valid, y_valid, reference=lgb_train)\n\nparams = {\n 'objective': 'binary',\n 'metric': 'auc',\n 'learning_rate': 0.05,\n 'num_leaves': 64,\n 'subsample': 0.8,\n 'colsample_bytree': 0.8,\n 'n_estimators': 1000,\n 'verbose': -1,\n 'random_state': 42,\n 'class_weight': 'balanced'\n}\n\nmodel = lgb.train(\n params, lgb_train, \n valid_sets=[lgb_valid],\n callbacks=[lgb.early_stopping(100), lgb.log_evaluation(100)]\n)\n\n# Valid AUC\nvalid_pred = model.predict(X_valid)\nprint(\"Valid AUC:\", roc_auc_score(y_valid, valid_pred))\n\n# TEST 예측\ntest_pred = model.predict(X_test)\n\n# Submission\nsubmission = pd.DataFrame({\n 'customer_id': test['customer_id'],\n 'target': test_pred\n})\n\nsubmission.to_csv('submission_lgbm_rfm.csv', index=False)\nprint(\"Submission 파일 저장 완료!\")\nprint(submission.head())","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2026-04-04T08:44:06.461218Z","iopub.execute_input":"2026-04-04T08:44:06.461550Z","iopub.status.idle":"2026-04-04T08:44:29.325888Z","shell.execute_reply.started":"2026-04-04T08:44:06.461522Z","shell.execute_reply":"2026-04-04T08:44:29.324630Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"# 제출용 파일 이름 생성 trial01\n\nsubmission = pd.DataFrame({\n 'customer_id': test['customer_id'],\n 'target': test_pred\n})\n\nsubmission.to_csv('submission.csv', index=False)\nprint(\"submission.csv 파일 생성 완료!\")\nprint(submission.head())","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2026-04-04T08:47:28.033178Z","iopub.execute_input":"2026-04-04T08:47:28.033524Z","iopub.status.idle":"2026-04-04T08:47:28.142546Z","shell.execute_reply.started":"2026-04-04T08:47:28.033496Z","shell.execute_reply":"2026-04-04T08:47:28.141447Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"import os\nprint(\"submission.csv 파일이 제대로 만들어졌는지 확인:\")\nprint('submission.csv' in os.listdir('/kaggle/working/'))\n\n# 파일 크기 확인\nif 'submission.csv' in os.listdir('/kaggle/working/'):\n size = os.path.getsize('/kaggle/working/submission.csv') / 1024\n print(f\"파일 크기: {size:.1f} KB\")","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2026-04-04T08:48:39.610059Z","iopub.execute_input":"2026-04-04T08:48:39.610843Z","iopub.status.idle":"2026-04-04T08:48:39.617516Z","shell.execute_reply.started":"2026-04-04T08:48:39.610809Z","shell.execute_reply":"2026-04-04T08:48:39.616304Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"# 제출용 파일을 정확한 이름으로 다시 생성\nsubmission = pd.DataFrame({\n 'customer_id': test['customer_id'],\n 'target': test_pred\n})\n\nsubmission.to_csv('submission.csv', index=False)\n\nprint(\"submission.csv 다시 생성 완료!\")\nprint(\"파일 크기:\", round(os.path.getsize('/kaggle/working/submission.csv') / 1024, 1), \"KB\")\nprint(submission.head())","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2026-04-04T08:51:21.312543Z","iopub.execute_input":"2026-04-04T08:51:21.312912Z","iopub.status.idle":"2026-04-04T08:51:21.418488Z","shell.execute_reply.started":"2026-04-04T08:51:21.312883Z","shell.execute_reply":"2026-04-04T08:51:21.417235Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"# submission.csv를 다시 정확히 생성 (Kaggle이 인식할 수 있게)\nsubmission = pd.DataFrame({\n 'customer_id': test['customer_id'],\n 'target': test_pred\n})\n\nsubmission.to_csv('submission.csv', index=False)\n\nprint(\"submission.csv 다시 생성 완료!\")\nprint(\"파일 크기:\", round(os.path.getsize('/kaggle/working/submission.csv') / 1024, 1), \"KB\")\nprint(submission.head())","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2026-04-04T08:55:58.982145Z","iopub.execute_input":"2026-04-04T08:55:58.983072Z","iopub.status.idle":"2026-04-04T08:55:59.089403Z","shell.execute_reply.started":"2026-04-04T08:55:58.983031Z","shell.execute_reply":"2026-04-04T08:55:59.088451Z"}},"outputs":[],"execution_count":null}]}