This work presents a new approach based on the use of stable dynamic models for dynamic data mining. Data mining is an essential technique in the process of extracting knowledge from data. This allows us to model the extracted knowledge using a formalism or a modeling technique. However, the data needed for knowledge extraction is collected in advance, and it can take a long time to collect. The objective is therefore to move towards a solution based on the modeling of systems using dynamic models and to study their stability. Stable dynamic models provide us with a basis for dynamic data mining. In order to achieve this objective, the authors propose an approach based on agent-based models, the concept of fixed points, and the Monte-Carlo method. Agent-based models can represent dynamic models that mirror or simulate a dynamic system, where such a model can be viewed as a source of data (data generators). In this work, the concept of fixed points was used in order to represent the stable states of the agent-based model. Finally, the Monte-Carlo method, which is a probabilistic method, was used to estimate certain values, using a very large number of experiments or runs. As a case study, the authors chose the evacuation system of a supermarket (or building) in case of danger, such as a fire. This complex system mainly comprises the various constituent elements of the building, such as rows of shelves, entry and exit doors, fire extinguishers, etc. In addition, these buildings are often filled with people of different categories (age, health, etc.). The use of the Monte-Carlo method allowed the authors to experiment with several scenarios, which allowed them to have more data to study this system and extract some knowledge. This knowledge allows us to predict the future situation regarding the building's evacuation system and anticipate improvements to its structure in order to make these buildings safer and prevent the greatest number of victims.
TopIntroduction
In recent years, the use of Data Mining techniques has expanded very rapidly, where it has become ubiquitous in the practices of businesses and individuals. The information explosion that is emerging in the world due to the size, sensitivity and complexity of the set of data collected; plays a leading role in the emergence of the field of data mining, where the use of data mining greatly improves the performance of data analysis techniques, in order to provide useful, understandable and up-to-date models or information. Additionally, data mining techniques and methods allow organizations to derive more information through understandable models that are constructed using homogeneous or heterogeneous data sets collected from various data sources (such as databases). distributed, web, data warehouses and satellite images).
Generally, data mining can be used in any field, where there is a need to analyze a large collection of data or to predict the evolution of a given process or system. Some fields of application for data mining include e-commerce, marketing, medicine, biology, information security, education and telecommunications.
The data mining process is based on data collections recorded in centralized or distributed storage media, where these collections are static and do not undergo any change. However, in real life, the analyzed data becomes more and more complicated and dynamic (continually changing), such as time series, real-time data, satellite data and the behaviors of individuals. In addition, the need to speed up the data mining process and to have up-to-date or real-time knowledge for, for example, security or competitiveness needs lead to the search for new tools to generate data collections., taking into account the characteristics of the system studied. The challenges mentioned above are mainly involved in the emergence of new trends in data mining, namely dynamic data mining, which consists of processing dynamic data, using dynamic models. Some of the techniques used in dynamic data mining include dynamic Bayesian networks, dynamic neural networks and dynamic clustering.
Data mining can have several trends depending on the types of data processed, such as text mining, web mining, and medical data mining. However, users of installed data mining systems are also interested in their associated technologies, and will be all the more so as most of these installations will have to be updated in the future as and when changes occur. can appear at the level of the systems studied, or in databases or data warehouses that are constantly changing over time.
Time series mining or data stream mining can be used by many applications such as e-commerce, intrusion detection, and ubiquitous data mining, which requires dynamic data mining models. For example, time series mining can be used to study cyclical trends, seasonal trends, random events or processes, relating to weather phenomena and stock market changes. On the other hand, in the mining of spatial, environmental or geographic data (which can have several aspects, such as distance, topology and aspect of time), the construction of data mining models which turns from synchronous with the studied system, greatly improves the quality and reliability of the information obtained.
Indeed, for each data mining technique used, we can find corresponding dynamic methods. For example, for Bayesian networks, there are dynamic Bayesian networks, for clustering, there is dynamic clustering and for classification, there is dynamic classification.
In addition, the data used for the data mining process is collected beforehand, and the collection time may be long. Since the main objective of data mining is to make the right decisions in order to gain in the operation of a system, leaving the system running for a long time in order to collect this data can lead to losses. . Therefore, there appears to be a need to speed up the data collection process.
Usually, the data gathered represents the history of a given system. However, to obtain these data more quickly, simulation tools can provide appropriate mechanisms and means. To achieve this goal, it is necessary to first build a simulator (a dynamic model) which properly represents the studied system, before exploiting the data generated by this alternative model.