Predicting defect density in software systems is essential for ensuring reliability. To streamline this process, managers aim to build prediction models to identify defective modules early, reducing testing costs and optimizing resource allocation. Feature reduction techniques play a pivotal role in enhancing defect prediction models by identifying critical features influencing defect occurrence. In this paper, we investigate defect density prediction using seven datasets obtained from the PROMISE repository. We apply seven feature reduction techniques, namely PCA, ANN, NLPCA, FastMap, Feature Agglomeration, TCA, and Random Projection, along with a novel stacking technique. We evaluate the performance of these techniques using three evaluation measures: Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Mean Absolute Error (MAE). Our analysis reveals that Random Projection and the proposed stacking technique consistently outperform other methods across all datasets. This study highlights the effectiveness of these techniques in predicting defect density and provides valuable insights for practitioners and researchers in software quality assurance.