A 15-step pre-processing procedure is proposed to improve the accuracy of sentiment mining of depression related posts in the tweets. Perhaps, for the first time, converting emoticons in the depression related tweets into text form is proposed during the pre-processing stage. In this paper, Term Frequency-Inverse Document Frequency with n-grams is used for feature extraction. Sentiment analysis conducted on a dataset consisting of 1.6 million depression related tweets using the proposed pre-processing module with feature extraction using Term Frequency-Inverse Document Frequency with n-grams and Logistic Regression (LR) for classification resulted in 81% of accuracy in detecting depression related tweets. © Springer Nature Switzerland AG 2020.