Utilização da metodologia KDD para descoberta de conhecimento em dados relacionados a toxicodependência e ideação suicida

Abstract

Suicidal ideation is an extremely important issue that affects all races and countries. Currently, there are gaps in studies on the application of the KDD methodology relating to the influence of drug addiction on suicidal ideation. The following work aims to discover knowledge in data using techniques for the prediction of suicidal ideation. For machine learning, the Decision Tree and Random Forest classification techniques were applied, and for data partition, the kfold method with 10 subsets. The database used is an American database belonging to the National Survey on Drug Use and Health (NSDUH), 5,055 instances and 19 attributes belonging to the demographic dimensions, mental health and drug addiction were explored, data from the year 2019 and 2020. Three experiments were carried out, all applying the two classification techniques in order to analyze the different behaviors of the model, alternating the dimensions of the attributes. The best performance was found in experiment 1 with the dataset containing all dimensions, applying the Decision Tree technique, which showed 63.7% accuracy, 55.2% precision and 49.1% recall. It was not possible to observe a considerable influence related to drug addiction in the dataset. In the majority of the experiments results, the technique that performed best was the Decision Tree. However, it was found that the percentage of correct prediction was still low, opening up opportunities for future work in the area with the analysis of new parameters, techniques and preprocessing of information that can improve prediction.