|
|
 |
| |
|
|
|
Predictive Analysis of Misuse of Alcohol and Drugs using Machine Learning Algorithms: The Case of using an Imbalanced Dataset from South Africa |
|
|
|
PP: 261-271 |
|
|
|
doi:10.18576/amis/170209
|
|
|
|
Author(s) |
|
|
|
Alexander Boateng,
Christopher Odoom,
Eric Teye Mensah,
Sarah Mensah Fobi,
Daniel Maposa,
|
|
|
|
Abstract |
|
|
| Drug and alcohol misuse has become a significant global distraction to the development of the youths and future leaders, especially in South Africa. The ideal solution to these unhealthy practices is for institutions, governments, and individuals to put preventive measures such as counselling, sanctions, and fines to help control drug and alcohol misuse. However, it is challenging to know the group of people or individuals engaged in such deteriorating activities by merely monitoring individuals physically. It makes it difficult even to plan the measures to control the issue. Therefore this paper has applied and compared six supervised machine learning algorithms to predict alcohol abuse and drugs across the nine provinces in South Africa to propose an ideal predictive model for detecting drug and alcohol misuse. The data used in the study was extracted from the 2019 General Household Survey conducted by Statistics South Africa, South Africa. The algorithms used in this paper include Random Forest, Naive Bayes, Support Vector Machines, Logistic Regression, Artificial Neural Networks, and Decision Tree. Results from the study identified Decision Tree to provide the highest recall of about 82.76 per cent, for alcohol abuse and drugs prediction, compared to the other five algorithms. In terms of the features and their importance, we found males across all the educational levels, mostly youth living in the Western Cape and Free State, to have played a vital role in the classification process. As part of the implication of the results in terms of policy formulation, we will edge the South African National Council on Alcohol and Drugs (SANCA) to draw up intervention programs to address issues of alcohol abuse and drugs, targeting all six attributes across all the provinces, especially Free State and North West provinces of South Africa. In conclusion, the approach used in this paper has effectively revealed an appropriate algorithm with a very high recall, reducing false negative, which means a successful reduction in misclassification of the important class. These results are reliable and valid for detecting drug and alcohol misuse at a low cost compared to other rigorous and demanding approaches.
|
|
|
|
|
|
|
|
