Machine Learning Approaches for Prediction of Laryngeal Cancer Based on Laboratory Test Results

Laryngeal cancer is approximately the twentieth most common cancer in the world with more than 150,000 new cases diagnosed annually. Laryngeal cancer, a prognostic serious disease associated with high mortality, is one of the most debilitating forms of cancer. Despite advances in therapy and novel surgical and non-surgical approaches, early diagnosis remains the best predictor of survival. Although cancer classification using gene expression data analysis has recently emerged in the research field, little is known of the relationship between pathology report results and final clinical results. In reality, vocal cord polyps are a common benign lesion, having the same voice disorder symptom as early laryngeal cancer. In this project, we use several popular machine learning techniques (logistic regression, random forest, PCA, etc.) to develop relevant prediction models to classify vocal cord polyps and early laryngeal cancer. The data set contains 63 variables for 5,000 patients. The k-fold cross-validation methodology is used in model evaluation and comparison. We compare the results from each method and provide some helpful instructions to support physician