Emotion detection from text plays a pivotal role in applications such as sentiment analysis, social media insights, and customer experience management. This study introduces a multi-model fusion approach for emotion detection using the Kaggle Emotion Text Dataset, a widely recognized benchmark that captures a variety of emotions across diverse textual inputs. The proposed framework employs a combination of machine learning classifiers, including Random Forest (RF), Logistic Regression (LR), Decision Trees (DT), Stochastic Gradient Descent (SGD) and Support Vector Machine (SVM). To maximize predictive performance, these models are integrated using two ensemble strategies: Stacking and Majority Voting. Stacking combines base models with a meta-classifier, enabling the system to learn intricate patterns in the data, while Majority Voting provides a simpler yet effective method for decision consolidation by leveraging collective model predictions. Performance evaluation is conducted using metrics such as accuracy, precision, recall, F-measure, False Positive Rate (FPR), and False Negative Rate (FNR). The results demonstrate that the Stacking approach achieves the highest accuracy of 99.92%, with precision of 99.68 %, recall of 99.19% and f-measure of 99.43%, respectively with Micro FPR of 0.0001, Micro FNR of 0.0007, Macro FPR of 0.0002 and Macro FNR of 0.0081. Majority Voting, while slightly less accurate, excels in reducing FPR and FNR, making it a valuable alternative in scenarios where minimizing misclassification is critical. This work underscores the potential of ensemble learning in addressing the complexities of emotion detection in text. The integration of diverse classifiers enhances prediction robustness and highlights the trade-offs between model complexity and real-world feasibility. By delivering a comprehensive evaluation and actionable insights, this single-author study contributes to advancing the field of emotion analysis and its practical applications.

Phishing threats continue to compromise online security by using deceptive URLs to lure users and extract sensitive information. This paper presents a method for detecting phishing URLs that employs optimal feature selection techniques to improve detection system accuracy and efficiency. The proposed approach aims to enhance performance by identifying the most relevant features from a comprehensive set and applying various machine learning algorithms, including Decision Trees, XGBoost, Random Forest, Extra Trees, Logistic Regression, AdaBoost, and K-Nearest Neighbors. Key features are selected from an extensive feature set using techniques such as information gain, information gain ratio, and chi-square (χ2). Evaluation results indicate promising outcomes, with the potential to surpass existing methods. The Extra Trees classifier, combined with the chi-square feature selection method, achieved an accuracy, precision, recall, and F-measure of 98.23% using a subset of 28 features out of a total of 48. Integrating optimal feature selection not only reduces computational demands but also enhances the effectiveness of phishing URL detection systems.