Given the substantial expansion of the airline sector in the last twenty years, it has become imperative to have efficient techniques for evaluating customer input in order to improve service provision. Using a variety of machine learning approaches on two separate datasets and their merged form, this research paper examines the effectiveness of sentiment analysis in extracting attitudes and emotions from airline service ratings. Initially, conventional machine learning methods including Random Forest, K-Nearest Neighbours (KNN), AdaBoost, Logistic Regression, Support Vector Machines (SVM), and Decision Trees were used. Among them, Random Forest demonstrated the best level of accuracy, reaching 99% on one of the available datasets. Following that, a hybrid model employing Convolutional Neural Networks and Bidirectional Long Short-Term Memory (CNN-Bi-LSTM) was investigated, achieving a maximum accuracy of 98%. The research subsequently progressed to utilizing transformer-based models such as DistilBERT, RoBERTa, ALBERT, Electra, and two versions of BERT (basic and big), with a specific emphasis on their ability to effectively process intricate contextual information. Despite comparable performance metrics among models, ALBERT demonstrated a little lower accuracy of 97% in a specific dataset, highlighting the intricate capabilities of transformer structures.