Making decisions for frost prediction in agricultural crops in a soft computing framework

Nowadays, there are many areas of daily life that can obtain benefit from technological advances and the large amounts of information stored. One of these areas is agriculture, giving place to precision agriculture. Frosts in crops are among the problems that precision agriculture tries to solve because produce great economic losses to farmers. The problem of early detection of frost is a process that involves a large amount of wheather data. However, the use of these data, both for the classification and regression task, must be carried out in an adequate way to obtain an inference with quality. A preprocessing of them is carried out in order to obtain a dataset grouping attributes that refer to the same measure in a single attribute expressed by a fuzzy value. From these fuzzy time series data we must use techniques for data analysis that are capable of manipulating them. Therefore, first a regression technique based on k-nearest neighbors in a Soft Computing framework is proposed that can deal with fuzzy data, and second, this technique and others to classification are used for the early detection of a frost from data obtained from different weather stations in the Region of Murcia (south-east Spain) with the aim of decrease the damages that these frosts can cause in crops. From the models obtained, an interpretation of the provided information is performed and the most relevant set of attributes is obtained for the anticipated prediction of a frost and of the temperature value. Several experiments are carried out on the datasets to obtain the models with the best performance in the prediction validating the results by means of a statistical analysis.