Skip to content

Files

Latest commit

 

History

History

Sales and Demand Forecasting

README.md

Business understanding:

“Global Mart” is an online store super giant having worldwide operations. It takes orders and delivers across the globe and deals with all the major product categories - consumer, corporate & home office.

Business Problem:

Now they wanted to finalise the plan for the next 6 months i.e. Need sales and demand forecast for the next 6 months, that would help them to manage the revenue and inventory accordingly. The store caters to 7 different market segments and in 3 major categories and wanted to forecast at this granular level, so the data had to be subseted into 21 (7*3) buckets before analysing the data. But not all of these 21 market buckets were of importance from the store’s point of view. So they wanted to found out the 2 most profitable (and consistent) segment from these 21 and forecast the sales and demand for these segments.

Data preparation:

  1. Segmented the whole dataset into the 21 subsets based on the market and the customer segment level.
  2. Convert the transaction-level data into a time series. Aggregate the 3 attributes - Sales, Quantity & Profit, over the Order Date to arrive at monthly values for these attributes.Then I got the three time-series
  3. Found the 2 most profitable and consistently profitable segments. Used the coefficient of variation for finding out the profitable segments among 21 market segments.
  4. Smoothen the data (require for classical decomposition).

Model building:

  1. Separate out the last 6 months values from the dataset
  2. Build the model using-

  • Classical decomposition

    • Fitted a multiplicative or additive model with trend and seasonality to the data, to find the global value of series
    • Subtracted the global values from the series
    • Modelled the residual series using the ARMA model, to get the local predictible series
    • Subracted the local predictible part from the residual series, we should be left with the noise
    • To check for the white noise performed ADF and KPSS tests

  • Auto ARIMA methodfor forecasting

    • Build a ARIMA model
    • Performed ADF and KPSS tests for checking white noise

Model evaluation:

  1. Forecasted the sales and quantity for the next 6 months using the above build models using MAPE.