Identification of habitual smokers through speech signal

Smoking is an addictive behavior and can result major health complications. Nowadays, many young adults tend to pick up this unhealthy habits which could potentially harm their health and affects the future workforce of the nation. Most of the habitual smokers have difficulties in ceasing this habit and require external assistance in the form of group therapy, medical interventions to quit smoking. Therefore, the main aim of this study is to investigate the speech signals of the subjects in an effort to identify the habitual smokers non-invasively. Through this detection, young smokers could be identified. Voice samples from VOice ICar fEDerico II from PhysioNet database were used for this study. Wavelet Packet Transform was used to extract non-linear features from the signals. Due to uneven data, ADASYN algorithm was used to produce a balanced dataset through synthetic data sampling. Subspace KNN and SVM classifiers were used for the investigations and classification accuracies up to 83.7% and 94% of AUC curve were observed from the analysis. The results suggests that the proposed method is effective in detecting habitual smokers, and can be considered as an early screening for smoking habits in young adults for targeted rehabilitation strategies.