Behavioral Predictive Framework User Guide
This user guide provides an overview over the GUI as well as all functional information about the service.
Via the sidebar you’re able to navigate through all the relevant sites. You can access the Data Management, Model Management and Result View. You also have access to logout, refresh the session, read the EULA again. In case you want more space on your screen you can also reduce the space of the sidebar with the button on the bottom.
The notifications are located on the right of the screen and contain the following information about actions triggered by a user:
- type: finished, error, running
- text about the details of the notification
- corresponding file or model
In order to use the BPF you have to provide your data to upload in a specific format and as a csv file. The file has to contain columns with the following informations and needs to be transactional, with one transaction representing one row in the file:
- identification: Person ID
- timestamp: a timestamp in datetime format corresponding to the transaction
- features/ target: a feature/ target column must be numeric and represents a userspecific event you want to select as a feature or a target(1: the feature occorurs at this timestamp, 0: the feature doesn’t occure)
You can upload new data and select the appropriate type. If you want to train a new model with your historic data, select type ‘input’. If you want to predict on newer data, select ‘score’.
Once you’ve uploaded your data successfully, you are able to edit your files. Besides restarting the meta extraction you can duplicate or the delete the file and change the title or description.
The Model Management site provides you with the proper tool in order to create a machine learning model with the data you’ve uploaded.
You choose between a general model in order to create a model, if your data matches the criteria described in the Data Management section. Custom models can be generated, this feature is currently not implemented yet, but will provide more accurate results based on company specific adjustments.
You are able to correct the datatypes calculated by the system via ‘correct data type’ and select the corresponding column types. The column types have to match with the dataset criteria mentioned earlier.
Once the model was succesfully trained, you can access detailed information about the loss and accuracy of the model. Furthermore a confusion matrix is calculated in order to represent the differences between predicted and actual data. Additionaly a ROC curve is created in order to see the true positive rate against the false positive rate.
After the model is trained you can score new data in order to predict on customer churn. In order to do that you choose a model of you choice with a scoring file of you choice. The features of the used input file during the model creation have to match. Once the data is scored, you are able to export, edit and delete the file.