Analysis of alcohol abuse using improved artificial intelligence methods

In this work, artificial intelligence (AI) based method has been proposed for analysis of alcohol abuse (AA) to evaluate drug risk and find an optimal method for analysis of AA. To detect alcohol abuse, twelve input features are selected. As the number of positive classes and negative classes are imbalanced, input features are processed using synthetic minority oversampling technique (SMOTE). The selected input features are given to the AI based method to predict whether an individual uses alcohol or not. Three AI methods (logistic regression, Naïve Bayes and gradient boost) have been used from which the most suitable method is chosen as optimal method for analysis of AA. Methods used for validation of the proposed AI based method are leave one out cross validation and test sample estimate. Among the three methods used, gradient boost method performs better than other method. Gradient Boost based AI method has 97.55% accuracy to predict alcohol user. Significant achievement of the proposed work is that the accuracy (nearly 97%) to detect alcohol abuse is much more than previously (nearly 70%) suggested methods. This is due to the efficient design of AI method implemented in this work. All the feature processing work and AI algorithm design work has been done using python software.