It seems that your use case can be addressed using Time Series models like AIRMA or LSTM, but it can get quite complex.
My recommendation would be to use Random Forest, a decision tree algorithm that can handle both numerical and categorical features such as issue date, invoice number, and amount. You can train the model on your historical payment data, and it would be able to predict payment dates for new invoices based on the input features.
Alternatively, you can use Neural Networks with Keras, and train the model to predict payment delay days based on input features. You may need to perform some feature engineering, especially since your input features contain datetime data. I suggest transforming these values into numbers that represent the difference in issue date vs payment date, etc.
I hope this helps. Let me know if you have any questions.