<![CDATA[Autism Spectrum Disorder (ASD), originally described by Leo Kanner in 1943, is a complex developmental condition that manifests through social, emotional, and behavioral challenges, often including speech delays and difficulties in interpersonal interactions. Despite significant advancements in diagnostic criteria over the years, accurate diagnosis of ASD in adults remains challenging due to limited access to comprehensive datasets and inherent methodological constraints. The Autism Screening Adult dataset used in this study exemplifies these issues, as it contains missing values and exhibits a marked class imbalance, both of which can adversely affect model performance. To address these challenges, we proposed a framework that integrates Random Forest classification with MissForest imputation and the Synthetic Minority Over-sampling Technique (SMOTE). MissForest effectively imputes missing data by employing an iterative random forest approach that preserves the underlying structure of the data without relying on strict parametric assumptions. Meanwhile, SMOTE generates synthetic samples for the minority class, thereby balancing the dataset and reducing prediction bias. Experimental evaluation through 10-Fold Cross Validation demonstrated that the application of SMOTE significantly enhanced model performance. Notably, the overall accuracy improved from 70.17% to 79.32%, and the AUC-ROC increased from 47.13% to 85.84%, indicating a robust improvement in the model’s ability to distinguish between positive and negative cases. These results underscore the critical importance of addressing data imbalance and missing values in predictive modeling for ASD. The promising outcomes of this study provide a solid foundation for developing more reliable diagnostic tools for adult ASD, and future research may further refine feature selection and incorporate additional data sources to optimize performance even further.]]>
