A comprehensive machine learning framework with particle swarm optimization for improved polycystic ovary syndrome (PCOS) diagnosis

Abstract

Abstract Polycystic Ovary Syndrome (PCOS) is a hormonal disorder primarily affecting women of reproductive age, characterized by irregular menstrual cycles, elevated male hormones, and ovarian cysts. Early detection and treatment are crucial to prevent long-term complications. This research utilizes clinical data from Kaggle to develop a non-invasive PCOS diagnostic system. The authors conducted comprehensive data preprocessing, feature engineering, and exploratory data analysis (EDA). The refined dataset was incorporated into various default machine learning (ML) algorithms, including LR, LDA, GNB, SVM, XGB, DT, AB, RF, and KNN, for PCOS classification with varying train test ratios 70:30 to 80:20. To further enhance the model’s performance, the authors hybridized all the ML models with Particle Swarm Optimization (PSO). Remarkably, the proposed LR+PSO model achieved the highest accuracy at 96.30%, demonstrating exceptional proficiency with an 80:20 train-test ratio. It significantly improved sensitivity to 94.44%, indicating enhanced detection of positive cases, all while maintaining the highest specificity at 97.22% and precision at 94.44% compared to other models. These results highlight a substantial improvement in integrated models, emphasizing the potential of this novel approach to enhance PCOS diagnosis in terms of accuracy and efficiency, ultimately benefiting individuals with PCOS in their treatment journey.