Application Of Data Mining Techniques In The Analysis Of Fire Incidents
Abstract
In recent years, the number of fire accidents decreased slowly, but the losses due to fire accident is still enormous every year. Under these circumstances, with the combination of data mining, data analysis and accident analysis theory, Microsoft SQL Server business intelligence components was used to perform data mining and implement the fire accident cases and data analysis with four commonly used algorithms - cluster analysis, association rules, time series and decision tree. Data mining is aiming to provide a reference on the application methods in analysis of incidents and has a certain academic value and practical significance in the analysis of the data processing and information on fire accidents.
Data Mining Technology
The mining system includes the data source (database and data warehouse), data collection and conversion server, data mining engine, Knowledge Base, knowledge assessments and visual user interface. The process is to first take the raw data acquisition and preparation, deposited in the database and data warehouse. Get the required data and after the conversion, in-depth learning through a variety of algorithms. Mining results after the evaluation and classification could help to find useful knowledge, and finally displayed to the output terminal.
Data mining Techniques and analysis of fire incident database
- Association rules analysis
- Clustering analysis
- Decision tree analysis
- Time series prediction
We must first create a mining model before fire accident case database mining. Field data parameter set mainly based on the consideration of predictability of data, integrity of the field, the actual meaning of the rules, the data of structural and other factors. For example: the date and time that fire broke out is quite random in the properties of fire accidents.
The mining structure modeling process of Clustering analysis is like the association rules. Clustering analysis is to classify each property, regardless of the value of accident property in actual requirements; each property will be given full consideration to the role of each attribute in the cluster. Therefore, trying to weed out the real meaning is not necessary, but properties that will impact the distinctive features of clustering results will be removed.
While setting the parameters of decision tree analysis, data that could not be involved in the decision tree classification need to be set to ignore (Ignore) to avoid the impact of the decision tree generation. Quarter and the month that fire accident happened were set to the input parameters (Input), the count of accidents, injury, death, loss (million) were set to the input and prediction (see fig. 5). A dependencies network (see fig. 6) was displayed after analysis to show the relations and the strength of link of different properties.
During the data model set-up process, the critical time (Key Time) property must be considered, because mining tools will be based on this property to determine the sequence of forecast data and data changes periodically. In this research, the date of the accident statistics was set as the Key Time value, to get better forecasting results. In the mining process, the number of injuries, death and loss (million) were set as the input and forecast fields, other fields were set as input parameters.
Conclusion
Four algorithms used in the fire accident data mining achieved the desired results. The data mining, to some extent, explores the ways and means in the application of fire accidents. Association rules produced in Association Rule analysis can help the accident analyst understand the relationship between accident attribute. Clustering Analysis groups the data fire accident and reflects the feature of different groups. Collecting determinant element of fire accident becomes easier through this analysis. This study, as a reference or a method, is valuable in fire accidents and other work safety accident data analysis. Data mining process can use different approach to analyze accident and explore many cases. In addition, the text data mining results need to be practiced further, but the performance of potential knowledge is still valuable considering that it’s hard to be obtained by traditional analysis methods; this knowledge can help policy makers understand the nature and essence of accident.
