How storieswithsiva/kaggle-predicting-future-sales works
Main Components
a) Data Pipeline Layer
• Input Data Sources (show multiple input files)
• sales_train.csv
• test.csv
• items.csv
• item_categories.csv
• shops.csv
b) Feature Engineering Layer
• Lag Feature Generator
• Mean Encoding Processor
• Holiday Feature Generator
• All-Pairs Generator (shop-item combinations)
c) Model Layer
• Primary Model: LightGBM
• Secondary Models:
• XGBoost Model 1
• XGBoost Model 2
• Ensemble Layer (optional)
d) Validation Layer
• Custom Time Series CV (6-fold)
• RMSE Evaluation
Important Details to Highlight
• Time series nature of the data (2013-2015)
• Monthly aggregation step
• Lag feature importance (1,2,3,5,12 months)
• Cross-validation strategy (blocks 28-33)