DATA-DRIVEN MUSHROOM EDIBILITY PREDICTION USING ML

Abstract

Mushrooms are widespread in many environments, but distinguishing edible varieties from poisonous ones is critically important. This study develops a machine learning classifier to predict whether a mushroom is edible or toxic based on observable physical traits. We utilize a publicly available dataset with 8,124 samples and 22 attributes describing each mushroom’s cap, gills, stalk, odour, and habitat. Several classification algorithms were evaluated, including Logistic Regression, Random Forest, Support Vector Machine (SVM), and K-Nearest Neighbours (KNN), using a hold-out validation approach. Random Forest, SVM, and KNN achieved the highest accuracies along with strong precision and recall, demonstrating their effectiveness for mushroom classification. The SVM model, chosen for its robust performance and efficiency, was deployed through a Streamlit web application. This user-friendly interface allows individuals to input a mushroom’s features or name and receive an immediate prediction of edibility versus toxicity.