A Tweets Mining Approach to Detection of Critical Events Characteristics using Random Forest

During a natural disaster, while most people are overwhelmed, governmental agencies are in charge of public safety, and they must timely provide true directions, by quick and ecient analysis of massive amount of information. To guaranty decision making process, it is needed to extract and organize most important information as soon as possible to avoid adding more confusion. In this paper we show an automatic analysis of complex data such as text generated during an emergency event. An unsupervised and recent popular machine learning model called random forest is trained to uncover and organize predominant features from a large set of tweets; providing a hierarchy of main variables which might indicate rules and an approximation of how information  ows during an unusual event. In our work we provide, rstly, a conversion process from text to numerical vectors for the training data; secondly, we introduce brie y random forest model; next, we expose an algorithm to adapt random forest to our problem; nally we show results for dierent con guration of the model and conclusions.