Meat consumption prediction: a data science prespective

The growth registered in the meat industry has presented some challenges. The price component, while revealing market trends, can be affected by external factors that cause fluctuations in other types of meat. The dissertation focused on an analysis over the period 1990-2022 of the evolution assumed by meat prices and for its different types: poultry, pig, bovine and ovine. A series of causes justified the oscillating behaviour of prices since the financial crisis of 2007-2008, the rise in the price of grains (a central element in animal feed) or government policies that restricted trade between countries. Projections were made for 5 years using different models, in order to determine which one had a better performance for predicting the meat price index. The comparison of the Prophet, SARIMAX and LSTM RNN models showed better overall accuracy in the Prophet. It stood out for its ease of implementation and adjustment of the different components, as well as the adaptation to strong seasonal patterns. When considering projections for the future, possible shocks in supply and demand, extreme weather conditions, exchange rate variability or health concerns must be taken into account. The complementary analysis on predicting pig meat consumption involved a causal analysis on the statistical inference of several variables and possible impacts on the increase/decrease in pig consumption. Variables such as mean years of schooling, GDP per capita and female labour participation showed a direct and significant relationship in the increase in consumption, as opposed to the religious component, which had an inverse impact. Additionally, approaches from different Machine Learning models (AdaBoost, Random Forest, Multi-Layer Perceptron) complementing the robust linear model (Huber Estimator) were combined to verify which was better in the predictability of pig consumption, with AdaBoost standing out from the others.