Research and Application of Improved Multiple Imputation Based on R Language in Fire Prediction

Abstract

An improved multiple imputation based on R language is proposed to deal with the miss of data in a fire prediction model, which can affect the accuracy of the prediction results. Hazard and operability (HAZOP) is used to accurately find the data related to the research purpose, and exclude data with a missing rate greater than 80% and small differences in characteristics. Then, by changing the m value in the mice package under the R language (R-mice), the relevant parameters of the complete filling factor set under different m values are obtained. The value of m is determined after observing and comparing the parameters. The proposed method fully considers the randomness of filling and the difference between the generated dataset. Taking Hubei Province as an example, the data processed by this method are used as the input of the Bayesian network, and the fire trend is used as the output. The results show that the improved multiple imputation based on R-mice can solve the problem of missing data very well, and have a high prediction effect (AUC = 94.0800). In addition, the results of the predictive reasoning and sensitivity analysis show that the government’s supervision has a vital influence on the trend of fires in Hubei Province.

Current Advances on the Assessment and Mitigation of Fire Risk in Buildings and Urban Areas—First Edition

“…In "Research and Application of Improved Multiple Imputation Based on R Language in Fire Prediction", Wang et al [3] propose an enhanced multiple imputation technique using R language for fire prediction models, focusing on addressing missing data that affect prediction accuracy. Their objective was to utilise Hazard and Operability (HAZOP) analysis to accurately identify and exclude data with substantial missing rates.…”