import streamlit as st import pickle import numpy as np import tensorflow as tf from tensorflow.keras.models import load_model from tensorflow.keras.preprocessing.sequence import pad_sequences # 1. SETUP VARIABLES max_length = 50 # 2. LOAD ASSETS (Cached so it only loads once) @st.cache_resource def load_model_and_tokenizer(): model = load_model('src/best_lstm_model.keras') with open("src/spam_tokenizer.pickle", "rb") as handle: tokenizer = pickle.load(handle) return model, tokenizer try: model, tokenizer = load_model_and_tokenizer() except Exception as e: st.error(f"Error loading model assets: {e}") # 3. STREAMLIT INTERFACE LAYOUT st.title("🚫 LSTM Spam Email Detector") st.write("Please type or paste an email below to check if it's Spam or Ham.") # Text Box for User Input user_message = st.text_area("Email Message", height=150, placeholder="Enter your text here...") # Analyze Button if st.button("Predict"): if user_message.strip() == "": st.warning("Please enter a message to analyze.") else: # Preprocess text input cleaned_message = str(user_message).lower() data_sequences = tokenizer.texts_to_sequences([cleaned_message]) data_padded = pad_sequences(data_sequences, maxlen=max_length, padding='post', truncating='post') processed_text = np.array([data_padded[0]]) # Run inference prediction = model.predict(processed_text) # Extract the exact single number from the array layer prediction_score = float(prediction[0][0]) # Extra validation for short texts: look for high-risk text indicators spam_keywords = ["won", "winner", "prize", "gift card", "claim", "usd", "cash", "urgent"] has_trigger_words = any(word in user_message.lower() for word in spam_keywords) # Decision Logic (Combines your 0.80 metric with trigger keyword validation) if prediction_score >= 0.80 or (prediction_score > 0.10 and has_trigger_words): # SPAM EMAIL FOLLOW THIS CONDITION st.error(f"🚨 **SPAM** (Confidence score: {prediction_score:.2%})") else: st.success(f"✅ **HAM / LEGITIMATE** (Confidence score: {prediction_score:.2%})") # Extract the exact decimal scalar score # prediction_score = float(prediction[0][0]) # Display output based on your 0.6 precision threshold # if prediction_score > 0.80: # st.error(f"🚨 **SPAM** (Confidence score: {prediction_score:.2f})") #else: # st.success(f"✅ **HAM / LEGITIMATE** (Confidence score: {prediction_score:.2f})")