Artificial neural networks for drug vulnerability recognition and dynamic scenarios simulation

FRANÇAIS :
112 variables (éducation, environnement, sexualité, trajectoire) non liées au comportement des toxicomanes, collectées auprès de 223 héroïnomanes et 322 non-toxicomanes, ont permis de tester la capacité de différents réseaux neuronaux artificiels (RNA) à distinguer les toxicomanes des non toxicomanes. Le réseau Supervised Feed Forward a correctement classé 95% de l'échantillon. Seules 47 variables sur 112 ont été sélectionnées mathématiquement pour "éduquer" le RNA. Elles ont permis de tester différents types de RNA : Back propagation avec Soft Max, Learning Vector Quantization, Logicon Projection, Radial Basis Function, Squashing, Fuzzy Art Map, réseau Modular Neural. Enfin, le réseau contrainte-satisfaction a été utilisé pour simuler des schémas dynamiques permettant de déterminer les variables provoquant le passage des non-toxicomanes vers la toxicomanie et vice-versa. (A partir du résumé des auteurs)
ENGLISH :
A collection of 112 variables not traditionally connected to drug user's behavior, collected from 223 heroin addicts and 322 non-users, were used to test the capability of different types of Artificial Neural Networks (ANNs) to classify the drug users and the non-drug users. The Supervised Feed Forward Network was abble to classify approximately 95% of the sample with accuracy. A special sensitivity tool selected only 47 among the 112 independent variables as necessary to train the ANN. Different types of ANN (Back Propagation with Soft Max; Learning Vector Quantization, Logicon Projection, Radial Basis Function, Squash, Fuzzy Art map, Modular Neural Network) were tested on the new 47 variables to decide which kind of ANN was better to classify the sample. A Constraint Satisfaction Network, specifically created by Semeion, was used to simulate a dynamic fuzzy map of the drug user's world and decide the fuzzy membership of a subject from the fuzzy membership of the drug users to the fuzzy membership of non-users and vice versa. (From authors'abstract)