import pandas as pd import numpy as np from statsmodels.tsa.arima.model import ARIMA import warnings warnings.filterwarnings('ignore') # Load Black Friday historical data dates = pd.date_range('2019-11-29', periods=24*4, freq='H') # 4 years of Black Friday np.random.seed(42) # Create realistic Black Friday pattern hourly_pattern = [] for day in range(4): # 4 years for hour in range(24): base = 1000 # Midnight online rush (12 AM - 2 AM) if hour in [0, 1]: sales = base * 3.5 # Early morning dormant (2 AM - 5 AM) elif 2 <= hour < 5: sales = base * 0.3 # Doorbuster rush (5 AM - 9 AM) elif 5 <= hour < 9: sales = base * 5.2 # Morning shopping (9 AM - 12 PM) elif 9 <= hour < 12: sales = base * 3.8 # Lunch lull (12 PM - 2 PM) elif 12 <= hour < 14: sales = base * 1.5 # Afternoon steady (2 PM - 6 PM) elif 14 <= hour < 18: sales = base * 2.5 # Evening surge (6 PM - 9 PM) elif 18 <= hour < 21: sales = base * 4.2 # Night wind down (9 PM - 12 AM) else: sales = base * 2.0 # Add noise and yearly growth sales *= (1 + day * 0.1) # 10% yearly growth sales += np.random.normal(0, 200) hourly_pattern.append(sales) sales_data = pd.Series(hourly_pattern, index=dates[:len(hourly_pattern)]) print("📊 BLACK FRIDAY HISTORICAL DATA") print("="*50) print(f"Date range: {dates[0].date()} to {dates[-1].date()}") print(f"Total hours: {len(sales_data)}") print(f"Average hourly sales: ${sales_data.mean():,.2f}") print(f"Peak hour sales: ${sales_data.max():,.2f}") # Fit ARIMA model print("\n🔧 TRAINING ARIMA MODEL") print("="*50) print("Finding optimal parameters...") # ARIMA(p, d, q) - p=autoregressive, d=differencing, q=moving average model = ARIMA(sales_data[:-24], order=(2, 1, 2)) fitted = model.fit() print("✅ Model trained successfully!") print(f"AIC Score: {fitted.aic:.2f}") print(f"BIC Score: {fitted.bic:.2f}") # Make predictions for next Black Friday (24 hours) print("\n🔮 PREDICTIONS FOR THIS BLACK FRIDAY") print("="*50) forecast = fitted.forecast(steps=24) actuals = sales_data[-24:] for hour, (pred, actual) in enumerate(zip(forecast, actuals)): time = f"{hour:02d}:00" accuracy = 100 - abs(pred - actual) / actual * 100 print(f"{time} - Predicted: ${pred:,.0f} | Actual: ${actual:,.0f} | Accuracy: {accuracy:.1f}%") # Calculate overall accuracy mae = np.mean(np.abs(forecast - actuals)) mape = np.mean(np.abs((forecast - actuals) / actuals)) * 100 print("\n📈 MODEL PERFORMANCE") print("="*50) print(f"Mean Absolute Error: ${mae:,.2f}") print(f"Mean Absolute Percentage Error: {mape:.1f}%") print(f"Overall Accuracy: {100-mape:.1f}%")

from prophet import Prophet import pandas as pd import numpy as np # Prepare data for Prophet prophet_data = pd.DataFrame({ 'ds': dates[:len(sales_data)], # Date column 'y': sales_data.values # Target variable }) # Add Black Friday as a holiday black_fridays = pd.DataFrame({ 'holiday': 'black_friday', 'ds': pd.to_datetime(['2019-11-29', '2020-11-27', '2021-11-26', '2022-11-25', '2023-11-24']), 'lower_window': -1, # Include Thanksgiving 'upper_window': 2, # Include weekend }) print("🔮 PROPHET MODEL - BUILT FOR BLACK FRIDAY") print("="*50) # Initialize Prophet with custom seasonality model = Prophet( changepoint_prior_scale=0.05, # Flexibility of trend seasonality_prior_scale=10, # Strength of seasonality holidays=black_fridays, # Include Black Friday effect yearly_seasonality=True, weekly_seasonality=False, # Not relevant for Black Friday daily_seasonality=True # Critical for hourly patterns ) # Add custom regressors for retail patterns prophet_data['is_doorbuster'] = (prophet_data['ds'].dt.hour >= 5) & (prophet_data['ds'].dt.hour < 9) prophet_data['is_online_rush'] = (prophet_data['ds'].dt.hour >= 0) & (prophet_data['ds'].dt.hour < 2) prophet_data['is_evening_surge'] = (prophet_data['ds'].dt.hour >= 18) & (prophet_data['ds'].dt.hour < 21) model.add_regressor('is_doorbuster') model.add_regressor('is_online_rush') model.add_regressor('is_evening_surge') print("📊 Training on historical Black Friday data...") model.fit(prophet_data[:-24]) # Make future predictions future = model.make_future_dataframe(periods=24, freq='H') future['is_doorbuster'] = (future['ds'].dt.hour >= 5) & (future['ds'].dt.hour < 9) future['is_online_rush'] = (future['ds'].dt.hour >= 0) & (future['ds'].dt.hour < 2) future['is_evening_surge'] = (future['ds'].dt.hour >= 18) & (future['ds'].dt.hour < 21) print("✅ Model trained with special Black Friday patterns!") # Generate forecast forecast = model.predict(future) print("\n🎯 BLACK FRIDAY HOURLY PREDICTIONS") print("="*50) predictions = forecast[['ds', 'yhat', 'yhat_lower', 'yhat_upper']].tail(24) for _, row in predictions.iterrows(): hour = row['ds'].strftime('%H:00') pred = row['yhat'] lower = row['yhat_lower'] upper = row['yhat_upper'] # Staffing recommendations based on predictions staff_needed = int(pred / 200) # 1 staff per 200 customers/hour print(f"{hour} - Sales: ${pred:,.0f} [{lower:,.0f}-{upper:,.0f}] | Staff needed: {staff_needed}") print("\n💡 KEY INSIGHTS") print("="*50) print("• Peak hour: 6:00 AM - Prepare maximum staff") print("• Online surge: 12:00 AM - Ensure website capacity") print("• Lunch lull: 12:00 PM - Reduce staff, prepare for afternoon") print("• Evening surge: 6:00 PM - Second wave preparation needed")