Ensuring high-quality medical text data is a persistent challenge in healthcare analytics, as issues such as inconsistency, incompleteness, and hidden anomalies can substantially compromise the reliability of clinical decision-making and downstream research outcomes. This article presents and assesses a hybrid isolation forest (IF)–support vector machine (SVM) model (HIFSVM) to improve the quality of medical text data via anomaly detection, comparing with IF and SVM. The research used three public health datasets: Alzheimer’s Disease and Healthy Aging Data (ADHAD), Breast Cancer Global Dataset, and 500 Cities: The Local Data for Better Health (500LDFB), and employed 10 extensive data quality metrics: accuracy, completeness, consistency, timeliness, validity, uniqueness, precision, change rate, error rate, and data density. The models’ anomaly detection performance was evaluated using synthetic ground truth, cross-validation, and expert validation. The HIFSVM model reported increased change rates (e.g., 0.96 for ADHAD’s data value and 1.00 for 500LDFB’s confidence limits), demonstrating considerable potential for tackling intricate or dynamic medical datasets that necessitate sophisticated anomaly detection. It also demonstrated enhanced detection accuracy (F1-scores: 0.89–0.91 across datasets), markedly surpassing IF (0.83–0.87) and SVM (0.74–0.79) (both p < 0.05). Validation established that 79–91% of the identified anomalies were clinically significant, exhibiting considerable inter-rater agreement (κ = 0.71–0.84). These findings demonstrate that HIFSVM proficiently detects authentic anomalies while maintaining robust data quality evaluation capabilities. The study provides a structured methodological framework for data preprocessing, feature engineering, and anomaly detection that healthcare professionals can employ to enhance the reliability of clinical decision-making, thereby improving the accuracy of clinical studies and ensuring better patient outcomes.