The Cox-Proportional Hazards model is introduced as a potential tool to understand customer behavior relating to churn or disconnections in the telecommunications space. An overview of Survival Analysis is provided along with its associated quantities and metrics with examples to better illustrate concepts. The derivation of the classical Cox-Proportional Hazards model is discussed in detail and applied to network behavioural data. The development of additive models and generalised additive models are traced and described as a prelude to the additive Cox-Proportional Hazards Regression.
The cubic splines are used as a tool to automatically detect trends in the customer data and this is compared to the findings of the classic Cox-Proportional Hazard.
It is shown that using the cubic splines, trends are automatically detected in the data and the cubic spline functions themselves can be easily derived and implemented using the RCS macro. Insights and recommendations are reported on and made available to the Network for use in informing future retention strategies and in general, to better understand customer specific behaviour.