CODEWITHSUNDEEP
pythonpandasmachine-learningregressionforecasting
pynewbee
pynewbee

Reputation: 679

Forecasting, (finding the right model)

Using Python, I am trying to predict the future sales count of a product, using historical sales data. I am also trying to predict these counts for various groups of products.

For example, my columns looks like this:

Date Sales_count Department Item Color

8/1/2018, 50, Homegoods, Hats, Red_hat

If I want to build a model that predicts the sales_count for each Department/Item/Color combo using historical data (time), what is the best model to use?

If I do Linear regression on time against sales, how do I account for various categories? Can I group them?

Would I instead use multilinear regression, treating the various categories as independent variables?

Upvotes: 1

Views: 382

Answers (1)

Sreekiran A R
Sreekiran A R

Reputation: 3421

The best way I have come across in forecasting in python is using SARIMAX( Seasonal Auto Regressive Integrated Moving Average with Exogenous Variables) model in statsmodel Library. Here is the link for a very good tutorial in SARIMAX using python Also, If you are able to group the data frame according to your Department/Item?color combo, you can put them in a loop and apply the same model. May be you can create a key for each unique combination and for each key condition you can forecast the sales. For example,

df=pd.read_csv('your_file.csv')
df['key']=df['Department']+'_'+df['Item']+'_'+df['Color']
for key in df['key'].unique():
    temp=df.loc[df['key']==key]#filtering only the specific group
    temp=temp.groupby('Date')['Sales_count'].sum().reset_index()
    #aggregating the sum of sales in that date. Ignore if not required. 
    #write the forecasting code here from the tutorial

Upvotes: 1

Related Questions