Intro

Thousands of companies around the world, from small startups to global corporations, find great value in being able to accurately predict sales, and it’s almost always one of the priorities for their Data Science / Analytics teams.

However, all of them seem to attempt to increase accuracy (reduce error)by focusing on mainly two things:

1) Feature engineering (getting the most out of your features)

2) Model/parameter optimization (choosing the best model & best parameters)

Both of the above are very necessary indeed, but there is a third thing that adds valuein a complementaryway, and it’s wildly underused not only in this use case but in most data science projects:

• Combining external information.

In this article, we’ll do a simple sales forecast model and then blend external variables (properly done).

What we’ll do

• Step 1: Define and understand Target
• Step 2: Make a Simple Forecast Model
• Step 3: Add Financial Indicators and News
• Step 4: Test the Models
• Step 5: Measure Results

Step 1. Define and understand Target

There have been several implementations of the popular Walmart Sales Forecast competition to predict their sales.

For this article, we’ll use the data and improve a model only by combining external information.

Walmart released data containing weekly sales for 99 departments (clothing, electronics, food…) in every physical storealong with some other added features.

Walmart dataset screenshot

For this, we will create an ML model with ‘_Weekly_Sales’ as target, _and train with the first 70% observations and test on the posterior 30%.

The objective is to minimize the Prediction erroron future weekly sales.

We’ll add external variables that impact or have a relationship with sales such as dollarindex, oilprice and newsabout Walmart.

We won’t use model/parameter optimization nor feature engineering so we can distinguish the benefit from adding the external features.

#nlp #forecasting #machine-learning #python #data-science