Clinical Laboratory Data-Based Bladder Cancer Prediction Using Machine Learning Approach

Bladder cancer (BC) is characterized by the abnormal growth of cells in the bladder’s tissues. While urethral cytology can detect BC early, its sensitivity is low, necessitating more accurate diagnostic methods. This research aimed to develop a machine learning-based BC classification model using clinical laboratory data. The study included 1,336 individuals with various types of cancer, including cystitis, BC, kidney cancer, uterine cancer, and prostate cancer. We employed a comprehensive data preprocessing pipeline followed by the application of eight diverse machine learning (ML) classifiers for the classification task. These classifiers included a Stacking Classifier, a Voting Classifier, a Decision Tree, a model tuned by Random Search, a Light Gradient Boosting Machine, a Multi-Layer Perceptron (MLP), a model tuned by GridSearchCV, and Gradient Boosting. The Stacking Classifier demonstrated superior performance in distinguishing BC from cystitis and other malignancies, achieving the highest accuracy (99.63%), precision (0.9963), recall (0.9963), F1 score (0.9963), specificity (0.9955), and AUC (1.0). This work highlights the potential of using clinical laboratory data with ML approaches to predict BC, offering a promising alternative to current diagnostic methods.