Detecting code smells is crucial for maintaining software quality and mitigating technical debt, as these subtle design issues complicate software maintainability and evolution. Existing heuristic-based tools and static analysis frameworks often struggle with the complexity of modern systems, yielding inconsistent and unscalable results. This research introduces a multiphase framework for code smell detection, integrating Modified Fuzzy C-Means with supervision (MFCMS) and Principal Component Analysis (PCA) to improve feature selection, reduce dimensionality, and enhance detection accuracy while ensuring scalability. The framework leverages MFCMS to address feature uncertainty and PCA to mitigate feature correlation, enabling efficient, interpretable feature selection that preserves essential data for accurate classification. Although the framework shows promise, challenges in generalizability to diverse datasets and code smells, as well as dataset imbalance, are acknowledged, offering directions for future research. This work advances the state of the art by providing a robust, scalable, and practical methodology for automated code smell detection, integral to modern software quality assurance practices.