mushroom-classifier

Asymmetric classification project to prioritize safety in mushroom foraging.

🎯 Goal

The goal of this project is to classify mushrooms as edible or poisonous. Because eating a poisonous mushroom is fatal, the model is optimized for 100% Recall (Safety First).

🛠️ Tools Used

• Python / Google Colab
• Scikit-Learn: RandomForest, CalibratedClassifierCV
• Pandas/Numpy: Data manipulation

📈 Methodology

• Calibration: Used Sigmoid calibration to ensure probability scores were accurate.
• Asymmetric Thresholding: Instead of the standard 0.5 threshold, I implemented a dynamic threshold calculated to catch every poisonous sample (Recall = 1.0).
• Categorical Handling: Used OneHotEncoding and treated missing values as 'unknown' to capture hidden toxicity patterns.

🏆 Competition Results

• The final model was deployed on unlabeled foraged data for a classroom competition.
• Achieved high edibility rates while maintaining absolute safety.