Comparing Traditional Machine Learning And Large Language Models: An Application To Mental Health Text Classification