Binary classification is a fundamental challenge in machine learning, with applications spanning finance, healthcare, and communications. Existing algorithms like Random Forest, Decision Trees, and Logistic Regression each have their strengths, but none consistently excel across diverse datasets. This paper introduces a novel approach that integrates modified stacking and voting ensemble techniques to improve the accuracy and robustness of binary classification. By enriching datasets with prediction probabilities and leveraging a soft voting classifier, the proposed method achieves a classification accuracy of 98.086% across multiple datasets. The results demonstrate significant improvements over traditional algorithms, offering a more effective solution for binary classification tasks. The findings hold promise for applications where accurate decision-making is crucial, such as medical diagnostics and fraud detection.