Machine learning algorithms for predicting smokeless tobacco status among women in Northeastern States, India

Use of smokeless tobacco (SLT) in women is very high and serious public health issue in the northeast states, India. Prediction on status of SLT use among women is a key to policy making and resource planning at district and community level in this region. This study aims to predict the status of smokeless tobacco use among women in northeast states of India by applying several machine learning (ML) algorithms. We used publicly available National Family Health Survey, 2015-16 data. Eight ML algorithms were used for the prediction on status of SLT use. Precision, specificity, sensitivity, accuracy, and Cohen's kappa statistic were performed as a part of the systematic assessment of the algorithms. Result of this study reveals that the best classification performance was accomplished with random forest (RF) algorithm accuracy of 79.51% [77.65-81.37], sensitivity of 87.75% [86.55-88.95], specificity of 65.19% [65.18-65.20], precision of 81.39%, F-measure of 84.35 and Cohen's Kappa was 0.545 [0.529-0.558]. It was concluded that the algorithm of random forest was found superior and performed much better as compared to the rest ML algorithms in predicting the status on smokeless tobacco use in women of northeast states, India. Finally, this research finding recommends application of RF algorithm for classification and feature selection to predict the status of smokeless tobacco as a core interest.