Development of an Improved Facial Analysis-based System for Predicting Drug Addiction using Random Forest Classification Algorithm

Abstract

Drug abuse has become a widespread global issue, affecting individuals, families, and communities. Machine learning techniques have shown promise in combating drug addiction prevalence through early prediction and immediate intervention. This study proposes an improved facial analysis-based drug addiction prediction system using a random forest classification algorithm, trained on facial images. Feature extraction and selection were performed using a histogram of oriented gradients and recursive feature elimination, respectively. The random forest classification model was tuned with grid search cross-validation, and evaluated using accuracy, precision, recall, and F1-score. Tuning the system significantly improved its performance, with accuracy increasing from 84.62% to 87.18%, precision from 82.61% to 83.33%, recall from 90.48% to 95.24%, and F1-score from 86.37% to 88.89%. This increase demonstrates the importance of hyperparameter tuning and the robustness of the random forest algorithm. Future studies can improve upon this work by incorporating a larger facial dataset for better practical results.