Anxiety disorder is now a common mental disorder worldwide. Its early prediction is a very complex process due to the clinical method. Here, the problem of delayed diagnosis has been solved by creating an early detection framework through a machine learning model, which is capable of detecting the level of anxiety. Although various models are used in ongoing studies, there is a shortage in achieving computational efficiency while maintaining high reliability on the dataset. We solved this problem by running 9 different machine learning models. Which range from conventional classifiers to more advanced gradient boosted ensembles. In our method, data pre-processing, encoding and class balancing have been done very carefully. As a result, the reliability of the said model can be ensured. Here, the results show that Random Forest gave more and better output than all other models. And gave 86.62% accuracy and highest precision, recall and F1 score output. Here MSE was only 13.38%. Then ensemble models like Support Vector Machine and Logistic Regresion also showed the second & third highest accuracy of 83.44%. The benefit of this study here is that gradient boosting frameworks like LightBGM are a very reliable solution for anxiety level detection. Which will help doctors in quick prediction and personalized treatment planning.