The concept of precision agriculture tries to integrate the problems of agriculture with new technologies, in order to provide effective, feasible and efficient solutions, always trying to obtain a higher yield from farmers. Agriculture is an area of high economic relevance in many places. Specifically in this study we will focus on the Spanish province of the Region of Murcia, where agriculture accounts for more than 20% of its economy. Frosts in crops are posing as a serious problem for farmers in this area due to climate change. In this paper we address the problem of frosts suffered by farmers in southeastern Spain with frosts in their crops. At the end of the winter season, temperatures vary by as much as 20 degrees Celsius from midday to night. These variations provoke the anticipation of the blossoming in stone fruit trees, having the risk of frosts at night.Thus, in this paper, the Intelligent Data Analysis have been applied to create predictive models of minimum temperature in plots. In addition, a selection process of the most relevant characteristics to predict the minimum temperature will be presented using the information provided by the models. The data necessary to carry out this study will be collected from the different weather stations of the Institute of Agricultural and Food Research and Development of Murcia. Specifically, data from forty weather stations have been studied, with the aim of finding local or global models that predict the temperature one hour in advance. The data analysis techniques used for the prediction models have been the M5 rule technique for predicting the minimum temperature and the C4.5 decision tree for classifying whether frost will occur or not. The results have identified the most relevant attributes both for predicting and characterize temperature and for classifying whether frost: dew point, vapour pressure deficit and maximum relative humidity occur. The results obtained indicate that both a local classification model and a local prediction model fit perfectly to the resolution of the problem obtaining on the one hand an error of less than 0.5 degrees Celsius for the prediction of the minimum temperature and on the other hand a precision of 98% for the classification of whether frost will occur or not.