PhD Research

System Evolution Analytics: Data mining and learning of Complex and Big Evolving Systems 

Project Members: Animesh Chaturvedi, Dr. Aruna Tiwari, Prof. Dave Binkley, Emeritus Prof. Nicolas Spyratos,  and Prof. Pietro Lio'

Usually, real-world evolving systems have many entities (or components), which evolves over time. A technique can be used to analyze the evolving system; the technique is named as System Evolution Analytics. Such techniques can be applied on an evolving system represented as a set of temporal networks. These techniques fall in the categories of system learning and system mining. The state series of an evolving system is denoted as SS = {S1, S2, . . . , SN}. Then, the connections (or relationships) between entities of each state are pre-processed to make a temporal network, and this resulted in a series of evolving networks EN = {EN1, EN2, . . . , ENN}. These temporal networks can be merged to make an evolution representor, which is used with learning and mining techniques for system evolution analysis. This made us to analyze evolving inter-connected entities of a system state series. The system learning is performed by applying active learning and deep learning on the evolution representor. The system mining is performed by applying two proposed pattern-mining techniques: network rule mining and subgraph mining. Specifically, the publications describe the following proposed approaches: System State Complexity, Evolving System Complexity, System Evolution Recommender, Stable Network Evolution Rule, and System Changeability Metric. The proposed approaches are used to generate recommendation and evolution information to perform system evolution analysis. For example, a graph theory application of a service change classifier algorithm assigning change labels to a web service’s call graph representing calls between operations and procedures, which helped to do Web Service Slicing by extracting a WSDL slice for Inter-operational analysis. 

Assuming some evolving systems change due to changing and evolving environment. We aim to introduce techniques that can be applied to discover hidden information about an evolving system. We are interested to develop algorithms and tools to mine and learn evolving system based on association rule mining, network motif mining, and deep learning. First, data mining technique like association rule mining identifies frequent patterns, associations, or correlations from evolving system. Second, network motifs are sub-graphs, which reoccur in a network by a particular pattern of interactions in an evolving system. Third, deep Learning is inspired from structure and function of the neural network in human brain. Deep learning forms Deep Neural Network (DNN), which is a kind of well-known Artificial Neural Networks (ANN). Fourth, efficient service evolution analytics using change/evolution mining based on change classification and evolution metrics. Fifth, Big Scholarly data analytics for the publication titles of 217 years in all research fields using parallel frequent pattern-growth algorithms for rules generation. 

Our data mining and machine learning techniques can identify hidden information about an evolving system. Such information is further helpful in decision-making. Our work aims to reduce human effort with the help of recommender/mining/learning tools construction. Evolving systems, Number of states used for the experiments, and their Internet links.

GitHub Repositories for 


Datasets


Journals

Conferences

System Network Analysis of an Evolving System, which provides summarized report (like time series, recommendations, rules etc.) to help in decision-making  and action-taking task .

Common Flow-Chart of all the approach system state series SS = {S1, S2...SN} to make set of evolving networks {EN1, EN2... ENN} , which produces mining and learning information that we aggregated to retrieve evolution information.