TB Treatment-Seeking Behavior

Ruixue Guo and Sze Suen CS229: Machine Learning, Dec 2012 Background For this project, we looked at a dataset containing tuberculosis (TB) symptoms, symptom onset time, and time until treatment was sought for 1691 non-urban households in Bihar, India. All households had at least one household member who was being treated for active TB disease. We are interested in identifying the relationship between an individual’s characteristics (such as disease symptoms an individual experienced) and the time it took for the individual to seek treatment. This could potentially guide health education programs that support timely identification of disease cases to prevent transmission. We first cleaned and prepared the data, explored some basic statistics to get an intuition for the dataset, then used several algorithms to see if we could accurately predict the time of seeking treatment, and, more importantly, use the algorithm outputs to identify reasonable characteristics of patients that sought care quickly. The methods we used were logistic regression, PCA, K-means clustering, and SVM. These results are presented below.