Detection of emergency situations using the Naive Bayes machine learning model.

Abstract

Currently, social networks have gained ground in generating and obtaining information instantly, this feature makes it very useful in the detection and warnings of emergencies such as road accidents, fires, storms, floods, etc. This has motivated the generation of a large number of works about the use of this information to face the problems generated by such emergencies, work such as that of A. Kansal, Y. Singh, N. Kumar and V. Mohindru "Detection of forest fire using Machine Learning technique" [1] or by Chamorro Verónica "Classification of tweets using supervised learning models" [2], show the use of machine learning techniques for the detection of extraordinary situations. After these catastrophic or emergency situations it is necessary to manage the services of attention and protection of the population, problems such as information chaos, uncertainty in the needs and services can find a solution in the timely detection of which events are really emergencies, so the purpose of this work we use X messages (Twitter) to classify which emergencies if they really are or are not. We use the machine learning algorithm known as Naive-Bayes in this problem of classification of X messages, to determine the real emergencies, with a result in the evaluation of the accuracy in the real emergency classification with a ratio of 73.4% among those classified as emergencies and classifies false emergencies with an accuracy of 75.4% among those classified as false. In general, the model obtained has an accuracy of 74.6% in its classification forecasts. It is considered that the use of a Naive-Bayes model for a prototype in the classification of emergency messages of the social network X could be very useful based on the results of the evaluation of its classification performance.