I am struggling on a university project in which the objective is to forecast sales of bank products.
The target variable is the amount of sales and besides the many numerical predictors I have only one categorical variable.
This categorical variable is nominal and includes all 172 different products. It is already at the grossest granularity available, hence, I cannot group them according to the product category they should theoretically belong to in order to reduce the possible values.
I know it is important to include this variable in the model but I feel it it wrong to create 171 dummy variables.
Some of you have ever experienced this issue? How did you proceed or how could you suggest me to approach this problem?
An idea I have is to cluster products based on the target variabe and then build one sales forecasting model for each cluster.