Machine Learning · FIB-UPC

Hypothyroid Classification

End-to-end binary classification pipeline — predict hypothyroidism from clinical data using 9 model families in sklearn. The interactive demos below run real trained models directly in your browser.

View Source on GitHub

ML Pipeline

FIB-UPC · Machine Learning·with Marta Granero

📂

ARFF data

hypothyroid.arff

🔧

Preprocess

Impute, scale, encode

📊

EDA

Correlation, distributions

🤖

Models

9 sklearn families

✅

Evaluate

CV, ROC, confusion matrices

Logistic RegressionLDA / QDANaive Bayesk-NNSVM (linear + kernel)MLP (neural nets)Random ForestRidge / LassoBayesian hyperparameter search

🏥

Hypothyroid predictor

Real LogReg model · runs in browser

Age50 years

TSH2 µU/mL

TT4109 nmol/L

T32 nmol/L

✅

Negative — healthy

Logistic Regression · 97.3% confidence

📍

k-NN on PCA projection

200 real test-set points · click to classify

k =

Click the plot to classify a point

Feature importance

Age

0.11

TSH

5.42

TT4

0.76

0.12

|coeff| from Logistic Regression on standardized features — TSH dominates prediction.

Dataset summary

hypothyroid.arff — UCI-style with numeric (age, TSH, T3, TT4, FTI, T4U) and categorical attributes. Target: binaryClass (P/N).

Challenges: heavy NaN, outliers in age, dropped TBG, class imbalance — handled via imputation, appropriate metrics (F1, ROC-AUC).

▸ Jupyter notebook preview

Open in Nbviewer ↗

▸ Run the notebooks locally

git clone https://github.com/cuberhaus/APA_Practica.git
cd APA_Practica
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
jupyter lab PracticaAPA-Hipotiroidismo-PolCasacubertaMartaGranero.ipynb