The invention provides a fire early warning method based on an improved naive Bayesian algorithm. The fire early warning method comprises the steps of determining fire characteristic attribute items and fire categories; acquiring a plurality of data and fire categories to which the data belong to form a training sample set and a test sample set respectively, and then further performing data preprocessing; calculating a prior probability according to the data of the training sample set, and calculating a conditional probability after smoothing processing based on a Laplace smoothing method; obtaining each fire characteristic attribute and a weighting coefficient of a characteristic attribute value, and combining prior and conditional probabilities to construct a dual-weighted naive Bayesianclassification model; compensating the prior probability, and improving a dual-weighted naive Bayesian classification model; and importing the data of the test sample set into an improved naive Bayesclassification model for classification to obtain a corresponding fire category. By implementing the method, the problem that in the prior art, the accuracy and implementation difficulty of a fire early warning algorithm still need to be improved is solved, and the fire prediction accuracy is improved.