Depression is considered to be among the most common mental disorders in the world and early detection is vital so as to ensure that with help it can be intervened and prevented with severe consequences such as suicide. Conventional methods of diagnosis are based on administering tests by clinicians and self-report measures which can be time-intensive and imprecise predictive measures. This paper introduces a machine learning-based system to detect and measure depression in patients using Patient Health Questionnaire-9 (PHQ-9) scale. The clinical data, which is comprised of reactions of psychiatric patients in a civil hospital, was used to train and test several machine learning models, including Light Gradient Boosting Machine (LightGBM), Random Forest, XGBoost, Decision Tree, and Support Vector Machine (SVM). Accuracy, macro F1-score, and precision, recall, and Cohen kappa coefficient were used to evaluate the performance of every model. The best model among the compared models was LightGBM, as the model had an accuracy of 97.62% and kappa of 0.9339, which was far much better than the other classifiers. Moreover, the system was also expanded to give severity-guided recommendations, including referring to the telepsychiatric services to severe cases, lifestyle to moderate cases, and supportive interventions to mild cases. The findings reveal that the combination of machine learning and PHQ-9 assessments can facilitate the process of depression screening, timely recommendations and clinical decision-making, particularly in healthcare environments with limited resources.