Having depression can have devastating effects on one’s mental as well as physical health. In addition to contributing greatly to the global burden of disease, depression is a leading cause of disability across the globe. An individual’s depression severity should be determined rather than just predicting whether he or she has depression. In this paper, a system to predict the level of depression using different machine learning multi-class classifiers is proposed for real-time dataset. The dataset is collected by asking questions individually which contains 392 data points and 9 features. The features are mainly about the changes in their life related to having depression. After data pre-processing and removing the noise and outliers, the best features are selected using the Step-Forward Feature Selection algorithm. The dataset is split into 80% and 20% for the train and the test set respectively. After this, Random Forest (RF), Logistic Regression (LR) classifier, Bagging Classifier (BC), and Naive Bayes (NB) were constructed using the selected best features. The robustness and performances of the classifiers are compared based on the accuracy, f1-score, and Mean Absolute Error (MAE) values. RF showed an accuracy of around 77%, BC around 70%. LR around 70% and NB around 69%. We found that RF gave better results than other classifiers based on not only its accuracy but also on its f1-score (0.725) which is the highest among other classifiers. RF also has the lowest MAE value (0.271). Finally, the weighted voting ensemble model is used for combining the performance of the classifiers. The accuracy for the ensemble model is 70% and the MAE value is 0.342. This paper is a preliminary report of the proposed system.