Subscription churn prediction
A churn prediction system includes at least one hardware processor, a memory including a historical sample set of subscriber data, and a churn prediction engine executing on the at least one hardware processor. The churn prediction engine is configured to identify the historical sample set, identify a set of attributes, automatically select a subset of attributes based on an information gain value, generate a decision tree by recursively generating nodes of the decision tree by computing an information gain value for each remaining attribute of the subset of attributes, identifying a highest attribute having the highest information gain value, and assigning the highest attribute to the node. The churn prediction engine is also configured to receive target data for a target subscriber, apply the target data to the decision tree, thereby generating a churn prediction for the target subscriber, and identify the target subscriber as a churn prediction.
1. A churn prediction system comprising:
at least one hardware processor;
an electronic memory storing instructions that when executed configure the at least one hardware processor to perform operations comprising:
identifying a historical sample set of subscriber data from a pool of historical samples stored in a database;
segmenting the historical sample set of subscriber data based at least in part on one or more of account age, subscription age, and gross merchandise value (GMV);
automatically selecting, using machine learning, a subset of attributes from a set of attributes of the segmented historical sample set of subscriber data based at least in part on an information gain value of each attribute of the set of attributes;
ranking each attribute in the subset of attributes using the machine learning;
generating a decision tree by assigning non-leaf nodes of the decision tree to corresponding attributes in the selected subset of attributes, each non-leaf node arranged in the decision tree according to the rank of the attribute assigned to the non-leaf node using the machine learning;
generating, using a programming language for analysis, a plurality of churn predictions for a plurality of target subscribers based at least in part on the decision tree and target data for the plurality of target subscribers stored in the database, each churn prediction of the plurality of churn predictions indicating a churn likelihood of a corresponding target subscriber of the plurality of target subscribers;
generating a ranked list of the plurality of target subscribers based at least in part on the plurality of churn predictions; and
causing presentation of the ranked list via a graphical user interface.
2. The churn prediction system of claim 1 , wherein the operations further comprise computing the information gain value of each attribute of the set of attributes, wherein automatically selecting the subset of attributes further includes selecting a pre-determined number of attributes having a highest information gain value.
3. The churn prediction system of claim 1 , wherein automatically selecting the subset of attributes further includes selecting attributes having information gain values above a pre-determined threshold.
4. The churn prediction system of claim 1 , wherein the operations further comprise:
receiving an indication of an analyst-identified attribute; and
adding the analyst-identified attribute to the selected subset of attributes for inclusion generating the decision tree.
5. The churn prediction system of claim 1 , wherein the operations further comprise:
identifying the historical sample set of subscriber data based at least in part on selecting the historical sample set of subscriber data from the pool of historical samples based at least in part on a distribution of a feature space of the pool of historical samples, wherein selecting the historical sample set of subscriber data further includes:
clustering the historical samples into K clusters; and
performing stratified subsampling of the historical samples using the K clusters as strata.
6. The churn prediction system of claim 1 , wherein generating the decision tree further includes biasing a leaf node as a survivor if a percentage of remaining samples at the leaf node labeled as churns are above a pre-determined threshold, the pre-determined threshold is higher than 50%.
7. A computer-implemented method for predicting churn comprising:
identifying a historical sample set of subscriber data from a pool of historical samples stored in a database;
segmenting the historical sample set of subscriber data based at least in part on one or more of account age, subscription age, and gross merchandise value (GMV);
automatically selecting, using machine learning, a subset of attributes from a set of attributes of the segmented historical sample set of subscriber data based at least in part on an information gain value of each attribute of the set of attributes;
ranking each attribute in the subset of attributes using the machine learning;
generating a decision tree by assigning non-leaf nodes of the decision tree to corresponding attributes in the selected subset of attributes, each non-leaf node arranged in the decision tree according to the rank of the attribute assigned to the non-leaf node using the machine learning;
generating, using a programming language for analysis, a plurality of churn predictions for a plurality of target subscribers based at least in part on the decision tree and target data for the plurality of target subscribers stored in the database, each churn prediction of the plurality of churn predictions indicating a churn likelihood of a corresponding target subscriber of the plurality of target subscribers;
generating a ranked list of the plurality of target subscribers based at least in part on the plurality of churn predictions; and
causing presentation of the ranked list via a graphical user interface.
8. The method of claim 7 , further comprising computing the information gain value of each attribute of the set of attributes, wherein automatically selecting the subset of attributes further includes selecting a pre-determined number of attributes having a highest information gain value.
9. The method of claim 7 , wherein automatically selecting the subset of attributes further includes selecting attributes having information gain values above a pre-determined threshold.
10. The method of claim 7 , further comprising:
receiving indication of an analyst-identified attribute; and
adding the analyst-identified attribute to the selected subset of attributes for inclusion generating the decision tree.
11. The method of claim 7 , further comprising:
identifying the historical sample set of subscriber data based at least in part on selecting the historical sample set of subscriber data from the pool of historical samples based at least in part on a distribution of a feature space of the pool of historical samples, wherein selecting the historical sample set of subscriber data further includes:
clustering the historical samples into K clusters; and
performing stratified subsampling of the historical samples using the K clusters as strata.
12. The method of claim 7 , wherein the decision tree includes a leaf node, wherein generating the decision tree further includes biasing the leaf node as a survivor if a percentage of remaining samples at the leaf node labeled as churns are above a pre-determined threshold, the pre-determined threshold is higher than 50%.
13. A non-transitory machine-readable medium storing processor-executable instructions which, when executed by a hardware processor, cause the hardware processor to perform operations comprising:
identifying a historical sample set of subscriber data from a pool of historical samples stored in a database;
segmenting the historical sample set of subscriber data based at least in part on one or more of account age, subscription age, and gross merchandise value (GMV);
automatically selecting, using machine learning, a subset of attributes from a set of attributes of the segmented historical sample set of subscriber data based at least in part on an information gain value of each attribute of the set of attributes;
ranking each attribute in the subset of attributes using the machine learning;
generating a decision tree by assigning non-leaf nodes of the decision tree to corresponding attributes in the selected subset of attributes, each non-leaf node arranged in the decision tree according to the rank of the attribute assigned to the non-leaf node using the machine learning;
generating, using a programming language for analysis, a plurality of churn predictions for a plurality of target subscribers based at least in part on the decision tree and target data for the plurality of target subscribers stored in the database, each churn prediction of the plurality of churn predictions indicating a churn likelihood of a corresponding target subscriber of the plurality of target subscribers;
generating a ranked list of the plurality of target subscribers based at least in part on the plurality of churn predictions; and
causing presentation of the ranked list via a graphical user interface.
14. The non-transitory machine-readable medium of claim 13 , wherein the processor-executable instructions further cause the hardware processor to compute the information gain value of each attribute of the set of attributes, wherein automatically selecting the subset of attributes further includes selecting a pre-determined number of attributes having a highest information gain value.
15. The non-transitory machine-readable medium of claim 13 , wherein automatically selecting the subset of attributes further includes selecting attributes having information gain values above a pre-determined threshold.
16. The non-transitory machine-readable medium of claim 13 , wherein the decision tree includes a leaf node, wherein generating the decision tree further includes biasing the leaf node as a survivor if a percentage of remaining samples at the leaf node labeled as churns are above a pre-determined threshold, the pre-determined threshold is higher than 50%.
17. The churn prediction system of claim 1 , wherein the machine learning is Waikato Environment for Knowledge Analysis (WEKA) machine learning.
18. The churn prediction system of claim 17 , wherein the machine learning uses an attribute evaluator of WEKA.