Forecasting sales for further 28 days for a given item of a store
Dataset: https://www.kaggle.com/c/m5-forecasting-accuracy/data
Copetition overview: https://www.kaggle.com/c/m5-forecasting-accuracy/overview
Dataset Overview:
The dataset consist of sales of previous 1941 days sales of 3049 items in 10 stores of 3 states in US. Apart from historical sales data we also have rate of each item at corresponding store and dates information like events on that corresponding date.
A customized metric known as WRMSSE based on MAPE is used as performance metric.
Performed 4 models on the dataset (Simpel Exponential Smoothing, XGBoostRegressor, CatBoostRegressor, LGBMRegressor).
EDA_FE.ipynb: Performed preprocessing and Exploratory Data Analysis on dataset and introduced lags and rolling features. Converted time series problem to supervised machine learning problem.
ses.ipynb: Performed simple Exponential smoothing.
models.ipynb: Performed all three above mentioned bossting algorithms.
final.ipynb: Final deployment model.
Scores:
Choosed CatBoostRegressor for final model.
Out of 5558 participants the ranks for score 0.685 were in range of 490-500's i.e the score can be considered under top 10% percentile rank.