Sentiment Analysis of Empathetic Dialogues- A Comprehensive Study on Machine Learning Algorithms and End-to-End Implementation

The current research project evaluates the efficiency and accuracy of diverse machine learning algorithms by employing the Empathetic Dialogues dataset, developed by the Hugging Face community for studying the underlying sentiments of different utterances. The sentiment classification accuracy was predicted using Random Forest Classifier, Voting Classifier, CatBoost Classifier, XGBoost Classifier, Extra Trees, Passive Aggressive Classifier, Support Vector Machines, Multi-Layer Perceptron Classifier, AdaBoost Classifier, LightGBM Algorithm, Decision Tree Classifier, Ridge Classifier, Gradient Boosting Classifier, Stochastic Gradient Descent Algorithm, k-Nearest Neighbors, Naive Bayes, and Logistic Regression. Confusion Matrix, ROC-AUC Curve and Precision-Recall (PR) Curve were employed to ensure model robustness, with test and training data being utilized to assure the models' reliability. From the analysis it is evident that the CatBoost Classifier attained the peak accuracy of 97.95%, succeeded by Voting Algorithm with 93.23% and Random Forest with 91.69%. k-NN and Naive Bayes Classifier recorded the lowest accuracies of 70.87% and 71.11%, respectively. The best-performing model was deployed on a Flask-based web application for real-time sentiment analysis, facilitating practical applications of the research. This paper provides valuable insights into developing accurate and robust sentiment analysis models with ML techniques.