Artificial intelligence to predict the price of olive oil

Artificial intelligence to predict the price of olive oil

2018/29/11 - Diego Hueltes, a Spanish computer engineer from Alcalá la Real (Jaén), has developed three models for predicting the price of olive oil using artificial intelligence.
"There are automatic learning algorithms that learn about past price fluctuations which, together with meteorological and production data, is capable of making an estimate close to reality," says to Mercacei this computer engineer specialized in Big Data and Machine Learning, a branch of artificial intelligence that serves, among other things, to create prediction models.

Hueltes wanted to make an investigation using the latest techniques of artificial intelligence and "it occurred to me that olive oil is a good topic, related to my land and I hope in the future it can be used and its benefits have repercussions in the province".

This computer engineer working in Marbella (Málaga) explains that there are external factors that affect and can not be predicted, but when the price depends solely on production, these algorithms are able to see the hidden relationships between the different variables to generate their estimates.

In this research, Hueltes has created several models and one of the "most interesting" is a model of "Deep learning" that simulates the behavior of neurons and the human brain to learn to predict, in this case, the price of extra virgin olive oil.

The engineer from Jaén has used data from the Andalusian Council on the price of olive oil in Jaén, the State Meteorological Agency (AEMET) and production of the Ministry of Agriculture, Fisheries and Food (MAPA).

Specifically, it has developed three models: one that predicts the price in the following week, another that predicts the price in four weeks and another that simply says whether the price will go up or down.

As detailed, there is a 2.9% absolute mean error for the prediction at four weeks; of 0.9% error for the prediction of the following week; and 76% accuracy in predicting the price direction. That 76% of success has been tested with the period between July 2017 and July 2018, with a 40% annual simulated benefit.

"The producer can provide the benefit of knowing what will happen in the short term with the price and thus plan the sales strategy. It could also be used to detect anomalies or changes in trends in order to react in time to sudden price changes," he adds.

Regarding how you can access this model, Hueltes advances that it will be an open source and free publication that can be found on its website www.hueltes.com, since "it is a scientific research that I do without the intention of profit".

The engineer from Jaen presents today for the first time this research at the congress Big Data Congress Lithuania and also tomorrow Friday November 30 at the congress Codemotion, held in Madrid.
Compartir en Google Bookmarks Compartir en Meneame enviar a reddit compartir en Tuenti