How it works

TruthRadar is powered by Flask (frontend), FastAPI (backend), and machine learning models using Scikit-learn, XGBoost, and more. It processes datasets like LIAR, FEVER, SciFact, and PHEME to detect misinformation in real time.

Datasets are cleaned — contractions expanded, punctuation and stopwords removed
Models are trained (Logistic Regression, Random Forest, XGBoost)
Vectorizers and models are serialized and served via FastAPI
Flask frontend queries FastAPI and displays predictions
ChatGPT summarizes the input, provides reasoning, and adds its own trust score

Models Utilized

LogisticRegression

A linear model that estimates class probabilities using a logistic function, commonly used for binary and multiclass classification.

BernoulliNB

A Naive Bayes variant optimized for binary/boolean features, assuming feature independence and using Bernoulli distributions.

RandomForest

An ensemble of decision trees that votes on outputs, offering strong performance and robustness to overfitting.

XGBoost

A gradient boosting framework that builds trees sequentially to correct errors, optimized for speed and accuracy.

PassiveAggressive

An online-learning linear classifier that updates only when it misclassifies, suitable for large-scale sparse data.

RidgeClassifier

A linear classifier that uses L2 regularization to prevent overfitting by shrinking weights.

SGDClassifier

A linear classifier trained with stochastic gradient descent, efficient for large-scale and streaming datasets.