Application of data mining techniques to predict the production of aflatoxin B1 in dry-cured ham.

Dry-cured ham may be contaminated with aflatoxin B 1 (AFB 1) produced by Aspergillus spp. Temperature and water activity (a w) are two key parameters that affect both ham ripening and AFB 1 production. The objective of this study was to predict AFB 1 production by Aspergillus parasiticus and Aspergillus flavus strains in conditions related to dry-cured ham ripening using data mining techniques. J48 decision tree, isotonic regression (IR), and multiple linear regression (MLR) were tested to (a) classify and predict AFB 1 concentration as a function of different days, temperatures and a w values and (b) predict the beginning of AFB 1 production as a function of different temperatures and a w values. For this, a model system based on a dry-cured ham-based medium was used. The percentage of correct classification was higher than 75%. R values to predict the concentration of AFB 1 when applying MLR were 0.81, being higher than those obtained after using IR. The models developed were validated with experimental data obtained after inoculating samples of dry-cured ham with two aflatoxigenic strains. The predicted AFB 1 concentration showed correlation coefficients ≥0.74 and prediction errors ≤0.38, confirming the feasibility of the prediction equations obtained. This information may help to make informed decisions to minimise the hazard posed by AFB 1 in dry-cured ham. • Data mining was applied to predict AFB 1 production by Aspergillus in dry cured ham. • A MLR model to predict aflatoxin risk in dry-cured ham was developed. • AFB 1 production was classified applying J48 decision tree. • The predictions were validated in dry cured ham with high correlation coefficients. • Data mining techniques could be used as predictive tools for AFs risk assessment. [ABSTRACT FROM AUTHOR]