Padding Interpolation, Median Imputation, RobustScalar, and Particle Swarm Optimization with Heterogeneous Classifiers: A Robust Combination for Effective Heart Disease Diagnosis

SANJAY DHANKA¹Ankur Kumar²Surita Maini³Nitin Kumar^1,4Jeewan Singh⁵Mudassir Khan^6*Mohamed Abbas⁶Amel Ksibi^7,8

• ¹Graphic Era Deemed to be University, Dehradun, India
• ²Indian Institute of Technology Mandi, Mandi, India
• ³Sant Longowal Institute of Engineering and Technology, Longowal, India
• ⁴Chitkara University Chitkara Design School, Rajpura, India
• ⁵Chandigarh University, Sahibzada Ajit Singh Nagar, India
• ⁶King Khalid University, Abha, Saudi Arabia
• ⁷Princess Nourah bint Abdulrahman University Central Library, Riyadh, Saudi Arabia
• ⁸Princess Nourah bint Abdulrahman University, Riyadh, Saudi Arabia

Heart disease (HD) remains a leading global cause of mortality, underscoring the critical need for early and accurate diagnosis. While machine learning (ML) offers promising avenues for automated diagnosis, many existing models suffer from data quality issues, suboptimal feature selection, and a lack of robustness. This study introduces a novel, robust diagnostic framework that synergistically combines advanced data preprocessing with an improved hybrid optimization-classification model. Our framework employs Padding Interpolation for handling missing values, Median Imputation for outlier correction, and RobustScalar for feature scaling to ensure data integrity. A key novelty is the introduction of an Improved Particle Swarm Optimization (IPSO) algorithm, which integrates a dynamic inertia weight and a mutation operator to enhance global search capability and prevent premature convergence. This IPSO is used for dual purposes: optimal feature selection and hyperparameter tuning of five heterogeneous classifiers (LR, LDA, GNB, SVC, XGBoost). Extensive experiments on a composite dataset from five public repositories demonstrate the superiority of our approach. The proposed Model 5 (IPSO-XGBoost) achieved state-of-the-art performance at a 90:10 train-test ratio, with an accuracy of 91.3%, sensitivity of 88.37%, specificity of 93.88%, precision of 92.68%, F1-score of 90.48%, and a remarkably high Diagnostic Odds Ratio (DOR) of 116.53. Statistical significance tests (p-value < 0.05) confirmed that the performance improvements over baseline models and existing methods are not due to random chances. The model also demonstrated consistent performance on independent Cleveland and Statlog validation datasets, proving its generalizability. This work establishes a comprehensive, robust, and highly effective pipeline for heart disease diagnosis.