Predicting when a customer will transact

[briantsmith90]

Hi, 

I have a business problem where I want to predict when a customer will transact by customer segmentation using historical transactions and want to know the best way to tackle the problem using the predictive tools. Essentially, the end result would tell me that customers in segmentation ABC will transact x% of the time and customers in segmentation XYZ will transaction x % of the time and so on.  There are 6 different customer segments. Below is fake data but the concept is the same 

[martinding]

Hi @briantsmith90 ,

This is an interesting problem. So how would you define "transaction x% of times"?

I see in the sample data, you have the # of transactions, this seems to form the numerator of the x% transaction ratio. 

One approach is to take the "# of transaction" as your target variable and try to predict this value for each segment and divide by your total number of transactions.

1. Take the # of transactions as your target variable that you are trying to predict.

2. Run a number of different classification models: logistic regression, random forest, naive bayes, boosting etc. and predict each row's (each customer's) transaction.

3. Aggregate the customer level transaction predictions -> groupby segment and sum the transaction predictions -> into segment level transactions.

4. Divide each segment level prediction by total number of transactions to get the transaction X% per segment.