Drinkers Voice Recognition Intelligent System: An Ensemble Stacking Machine Learning Approach

Abstract

Alcohol's dehydrating effects can cause vocal cords to dry out, potentially causing temporary voice changes and increasing the risk of vocal strain or damage. Short-term changes in pitch, volume, and alcohol consumption can cause voice clarity, which typically returns to normal after the effects of alcohol have subsided. Data science improves voice recognition by analyzing large volumes of voice data, training machine learning (ML) models, extracting meaningful features, and using deep learning and natural language processing techniques. The research paper proposes a novel approach for identifying and classifying individuals as drinkers or non-drinkers based on their voice patterns. We collect voice data from both drinkers and non-drinkers. The study utilizes an ensemble ML technique known as stacking to combine the predictive power of multiple models, including Naive Bayes, K-NN(Nearest Neighbors), Decision (DTS) Trees, and Support (SVM) Vector Machine. Different metrics, like AUC, CA, F1 score, Recall, and precision, are implemented to evaluate the performance of each model. The stacking ensemble model stands out with the highest AUC of 0.9890, showing its excellent capability to distinguish between individuals who drink and those who don't. The SVM model also performs exceptionally well, with an AUC of 0.9861. The study shows the efficacy of the ensemble ML approach for identifying voice-based drinkers, offering significant insights for creating intelligent systems to detect alcohol-related voice issues accurately. This research advanced ensemble Stacking ML techniques in alcohol use disorder detection and opened possibilities for developing real-world applications in healthcare and behavioral analysis.