Comparative Study of Machine Learning and Deep Learning Models for Early Prediction of Ovarian Cancer

Early detection of ovarian cancer is critical for improving patient survival and enabling timely therapeutic interventions. This study presents a comprehensive comparative analysis of classical machine learning and deep learning models for the prediction of ovarian cancer risk using a high-dimensional, multi-modal dataset comprising 200,100 patient records collected from January 2019 to December 2024. The dataset integrates clinical, genetic, imaging, reproductive, hormonal, and demographic features, including biomarkers such as CA-125, BRCA mutations, SNPs, miRNA levels, radiomic tumor characteristics, and patient lifestyle indicators. After preprocessing and robust feature scaling, the top 30 most informative predictors were selected using Random Forest-based importance scores. Classical models, including K-Nearest Neighbors, Support Vector Machines, Random Forest, and logistic regression, were trained and evaluated alongside ensemble approaches like Bagging and Stacking. Deep learning architectures, specifically an improved feedforward neural network (FNN) and a deep artificial neural network (ANN) implemented via MATLAB’s patternnet and trainNetwork, were optimized for multi-class risk prediction. Comparative results demonstrated superior performance of SVM and Random Forest models (~93–94% accuracy), while deep learning models achieved competitive accuracy (~78–83%). The findings highlight the potential of hybrid predictive frameworks leveraging multi-modal patient data for early ovarian cancer detection, offering a robust foundation for personalized risk stratification and precision medicine applications.

Keywords: Ovarian Cancer Prediction, Machine Learning, Deep Learning, Multi-Modal Patient Data, Early Detection.