The novel coronavirus disease (COVID-19) is an ongoing pandemic with large global attention. However, spreading false news on social media sites like Twitter is creating unnecessary anxiety towards this disease. The motto behind this study is to analyses tweets by Indian netizens during the COVID-19 lockdown. The data included tweets collected on the dates between 23 March 2020 and 15 July 2020 and the text has been labelled as fear, sad, anger, and joy. Data analysis was conducted by Bidirectional Encoder Representations from Transformers (BERT) model, which is a new deep-learning model for text analysis and performance and was compared with three other models such as logistic regression (LR), support vector machines (SVM), and long-short term memory (LSTM). Accuracy for every sentiment was separately calculated. The BERT model produced 89% accuracy and the other three models produced 75%, 74.75%, and 65%, respectively. Each sentiment classification has accuracy ranging from 75.88-87.33% with a median accuracy of 79.34%, which is a relatively considerable value in text mining algorithms. Our findings present the high prevalence of keywords and associated terms among Indian tweets during COVID-19. Further, this work clarifies public opinion on pandemics and lead public health authorities for a better society.