📈 Amazon Sales Forecasting Using Machine Learning

This project focuses on forecasting monthly sales for Amazon using a combination of time series analysis and ensemble machine learning techniques. Accurate sales predictions are critical for e-commerce businesses to improve inventory planning, reduce operational costs, and make informed marketing and budgeting decisions.

🔍 Project Overview

• Goal: Predict monthly sales using historical data to support business decision-making.
• Techniques Used: Time series preprocessing, feature engineering, ARIMA/SARIMA/Prophet models, ensemble learning (Extra Trees, XGBoost), and stacking.
• Performance: Achieved an R² score of 0.99 and 100% forecast accuracy within a 30% tolerance using a stacked regression model.

🧰 Key Features

• Data preprocessing: Handling missing values, resampling, outlier treatment
• Feature engineering: Lag features, rolling means, seasonal encoding (sin/cos)
• Time series models: ARIMA, SARIMA, Prophet
• Machine learning models: Extra Trees Regressor, XGBoost
• Model evaluation: MAE, RMSE, R² Score, custom accuracy within tolerance thresholds
• Forecast visualization: Plots with confidence intervals and future sales predictions

📊 Use Cases

• Demand forecasting for e-commerce platforms
• Inventory optimization
• Marketing campaign planning
• Business trend analysis

🛠️ Tech Stack

• Python (Pandas, NumPy, Matplotlib, Seaborn)
• Scikit-learn
• XGBoost
• Statsmodels (ARIMA/SARIMA)
• Prophet (by Meta)