Predictive Analytics, which in a nutshell is about predicting future outcomes based on past behavior. It is now routinely used in different industries, e.g., Retailers use it to predict customer buying patterns, airlines use it to set ticket prices reflecting past travel trends.
Predictive Network analytics is about using historical network data to predict network behavior. It can be used in an enterprise or service provider environments in:
Network operations – To predict network and hardware failures, Capacity planning
Security operations – To predict threats and attacks
Customer operations – To predict customer usage patterns
What is the relationship between Predictive and other types of Data Analytics?
Descriptive analytics provides summary statistics for the network for ab given period – e.g., hosts that consumed the most bandwidth in the past month – and visualizations of past data trends through charts and graphs. While this is still very useful in summarizing the network behavior and providing a network baseline, it requires engineers to sort through the data to find patterns.
Diagnostic analytics, which include accurately identifying the origin of the network faults and event correlation to minimize the number of alerts. The diagnosis process is referred by various names including Root cause analysis, Fault diagnosis or Fault localization. Identifying the causality of the network faults and correlating them requires extensive domain knowledge, the context of the current and historical faults. This seems to be the focus of Gartner’s AIops.
Predictive analytics, which attempts to predict what could happen by identifying patterns in the current and historical data. From a network and security perspective, perspective, this includes predicting which faults may occur next and estimating time-to-failure, forecasting traffic patterns and potential security threats.
How does Predictive analytics differ from Machine Learning?
Predictive Analytics and Machine Learning are used interchangeably because most predictive analytics solutions now use Machine Learning to do the predictions. However, it is not strictly necessary for predictions to be derived using ML.
All predictive analytics applications involve the following:
Data: The effectiveness of the predictions is directly proportional to the quality of the data it processes.
Statistical algorithms: Statistical techniques like linear and logistical regression used to derive data relationships and inference.
Assumptions: The assumptions regarding the historical data and their efficacy in predicting future outcomes.
Machine learning in addition to using historical data and statistical algorithms have some differences from non-ML based predictive analytics solutions. It can learn from features without specifying them explicitly and can recalibrate models automatically by doing pattern recognition and incremental learning. Meanwhile, predictive analytics works on parametrized data and must be refreshed with changed data.