System and method for detecting and forecasting the behavior of a system

A performance analysis system for detecting and forecasting a performance degradation of an M-System, the performance analysis system comprising:

• a memory;
• a processing unit configured for being connected to the M-System, said processing unit configured for automatically collecting data regarding the M-System during a period of time P1;
• a database for storing collected data;
• said processing unit further configured for analyzing the collected data collected during the period of time P1 by means of an elaboration routine having two distinct elaboration phases, respectively:
• a first elaboration phase, wherein the performance analysis system is configured for selecting among the collected data a set of data to be analyzed and removing any noisy data from the set of data;
• a second elaboration phase, wherein the performance analysis system is configured for correlating, for the set of data, data belonging to different sources, but generated at a same time within the period of time P1 in order to create M-System state data snapshots characterizing the M-System at different times Ti within the period of time P1;
• said processing unit being further configured for automatically generating input datasets from the M-System state data snapshots obtained at different times Ti and for creating a model for the M-System from the input datasets, wherein said model is created by training a predefined machine-learning algorithm by means of the input datasets, the predefined machine-learning algorithm being stored in said memory of the performance analysis system;
• said processing unit is further configured for starting a runtime phase after the period of time P1, wherein the runtime phase comprises:
• automatically collecting the data regarding the M-System during another period of time called P2 configured for starting after the period of time P1 in order to generate further M-System state data snapshots characterizing the M-System state at different times Tj within the period of time P2;
• and
• using the further M-System state data snapshots as an input for the model for automatically predicting the performance degradation of the M-System, and/or for optionally classifying a behavior of the M-System.