app.py

import gradio as gr
import re
from datasets import load_dataset
from sklearn.pipeline import make_pipeline
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score

# 1. Load dataset
dataset = load_dataset("ucirvine/sms_spam", split="train")
texts = dataset["sms"]
labels = [1 if label == "spam" else 0 for label in dataset["label"]]

# 2. Clean text
def clean_text(text):
    text = text.lower()
    text = re.sub(r"\W+", " ", text)
    return text.strip()

texts_cleaned = [clean_text(t) for t in texts]

# 3. Train/test split (use stratified sampling!)
X_train, X_test, y_train, y_test = train_test_split(
    texts_cleaned, labels, test_size=0.2, random_state=42, stratify=labels
)

# 4. Build model: TF-IDF + Logistic Regression
model = make_pipeline(
    TfidfVectorizer(ngram_range=(1, 2), stop_words="english", max_df=0.9),
    LogisticRegression(max_iter=1000, class_weight="balanced")
)
model.fit(X_train, y_train)

# 5. Show validation accuracy
y_pred = model.predict(X_test)
print("Validation Accuracy:", accuracy_score(y_test, y_pred))

# 6. Prediction function
def predict_spam(message):
    cleaned = clean_text(message)
    pred = model.predict([cleaned])[0]
    prob = model.predict_proba([cleaned])[0][pred]
    label = "🚫 Spam" if pred == 1 else "📩 Not Spam (Ham)"
    return f"{label} (Confidence: {prob:.2%})"

# 7. Gradio UI
iface = gr.Interface(
    fn=predict_spam,
    inputs=gr.Textbox(lines=4, label="Enter your SMS message"),
    outputs=gr.Text(label="Prediction"),
    title="📬 Improved SMS Spam Detector",
    description="Detects spam in SMS messages using Logistic Regression with TF-IDF bi-grams. Now with higher accuracy!"
)

if __name__ == "__main__":
    iface.launch(share=False)