[🇧🇷 Português] [🇺🇸 English]

6- Social Buzz AI - Fake News Detection Using Machine Learning

Fake.News.Detection.-.Machine.Learning.in.the.Fight.Against.Disinformation.mov.mov

Slide Presentation

Course: Humanistic AI & Data Science (4th Semester)
Institution: PUC-SP
Professor: Erick Bacconi

Tip

This repository 2-social-buzz-ai-GBoost-and-LowDefault-Modeling is part of the main project 1-social-buzz-ai-main. To explore all related materials, analyses, and notebooks, visit the main repository

• 1-social-buzz-ai-main Part of the Humanistic AI Research & Data Modeling Series — where data meets human insight.

• Access: NLP - Class 1 Repo

• Access Code

• Access True Dataset

• Access Fake Dataset

1. Introduction

• Fake news are false information mainly spread on social networks, which can cause serious political, social, and public health impacts.
• This study aims to apply Machine Learning (ML) algorithms to automatically detect fake news, offering a technological alternative to address this issue.

2. Study Objectives

• Test and compare different ML algorithms for detecting fake news.
• Assess the performance of each model in accuracy, sensitivity, and specificity.
• Propose an automated, replicable, and useful solution for society.

3. Detailed Methodology

3.1. Dataset

• Data obtained from Kaggle:
  • Fake: 23,481 samples
  • True: 21,417 samples
• Main columns: title, text, topic, date

3.2. Tools Used

Python + libraries: Pandas, NumPy, Matplotlib, Seaborn, Scikit-learn, NLTK

3.3. Data Processing

import pandas as pd
import numpy as np
import string
import nltk
from nltk.corpus import stopwords

# Load fake and true datasets
fake = pd.read_csv('Fake.csv')
true = pd.read_csv('True.csv')
fake['target'] = 1
true['target'] = 0

# Concatenate and shuffle records
data = pd.concat([fake, true], ignore_index=True)
data = data.sample(frac=1).reset_index(drop=True)

# Remove title and date columns
data.drop(['title', 'date'], axis=1, inplace=True)

# Clean text (lowercase, no punctuation, no stopwords)
nltk.download('stopwords')
stop_words = set(stopwords.words('portuguese'))

def clean_text(text):
    text = text.lower()
    text = text.translate(str.maketrans('', '', string.punctuation))
    text = " ".join([word for word in text.split() if word not in stop_words])
    return text

data['text'] = data['text'].apply(clean_text)

3.4. Visualization: WordCloud

from wordcloud import WordCloud
import matplotlib.pyplot as plt

wc = WordCloud(width=800, height=400, background_color='white').generate(' '.join(data['text']))
plt.figure(figsize=(10, 5))
plt.imshow(wc, interpolation='bilinear')
plt.axis('off')
plt.show()

3.5. Vectorization and Split

from sklearn.model_selection import train_test_split
from sklearn.feature_extraction.text import TfidfVectorizer

X = data['text']
y = data['target']

vectorizer = TfidfVectorizer(max_features=5000)
X_vect = vectorizer.fit_transform(X)

X_train, X_test, y_train, y_test = train_test_split(X_vect, y, test_size=0.2, random_state=42)

4. Applied Algorithms

Five supervised models were trained and evaluated:

Model	Accuracy	Remarks
Logistic Regression	98.92%	High precision, confusion matrix shows low error rate.
Decision Tree	99.6%	Best overall performance and lowest error.
Random Forest	98.74%	Good performance, consistent confusion matrix.
Support Vector Machine	99.5%	Excellent accuracy and precision, robust text model.
K-Nearest Neighbors (KNN)	60.84%	Low performance, high number of false negatives.

Metrics computed via confusion matrix (including TP, TN, FP, FN) and specific precision and sensitivity values for each model.

Example: Simplified Confusion Matrix (Decision Tree)

• True Positives (TP): 4,711
• False Positives (FP): 13
• False Negatives (FN): 22
• True Negatives (TN): 4,234

4.1. Model Training and Evaluation

Below are examples of models tested and evaluated.

Logistic Regression

from sklearn.linear_model import LogisticRegression
from sklearn.metrics import classification_report, confusion_matrix

lr = LogisticRegression()
lr.fit(X_train, y_train)
y_pred = lr.predict(X_test)
print(confusion_matrix(y_test, y_pred))
print(classification_report(y_test, y_pred))

Decision Tree

from sklearn.tree import DecisionTreeClassifier

dt = DecisionTreeClassifier()
dt.fit(X_train, y_train)
y_pred = dt.predict(X_test)
print(confusion_matrix(y_test, y_pred))
print(classification_report(y_test, y_pred))

Random Forest

from sklearn.ensemble import RandomForestClassifier

rf = RandomForestClassifier()
rf.fit(X_train, y_train)
y_pred = rf.predict(X_test)
print(confusion_matrix(y_test, y_pred))
print(classification_report(y_test, y_pred))

SVM

from sklearn.svm import SVC

svm = SVC()
svm.fit(X_train, y_train)
y_pred = svm.predict(X_test)
print(confusion_matrix(y_test, y_pred))
print(classification_report(y_test, y_pred))

KNN

from sklearn.neighbors import KNeighborsClassifier

knn = KNeighborsClassifier()
knn.fit(X_train, y_train)
y_pred = knn.predict(X_test)
print(confusion_matrix(y_test, y_pred))
print(classification_report(y_test, y_pred))

Confusion Matrix Visualization

import seaborn as sns

cm = confusion_matrix(y_test, y_pred)
sns.heatmap(cm, annot=True, fmt='d', cmap='Blues')
plt.xlabel('Predicted')
plt.ylabel('Actual')
plt.show()

5. Evaluation Metrics

• Accuracy: Overall model success rate.
• Precision: How well fake news are detected.
• Sensitivity: Model's ability to correctly identify actual fake news.
• Specificity: Model's ability to correctly identify actual real news.

Model Performance

Model	Precision	Sensitivity	Specificity
Logistic Regression	98%	99%	98%
Decision Tree	98.5%	99%	99%
Random Forest	98.5%	99%	98%
SVM	99%	99%	99%
KNN	99%	57%	19%

6. Detailed Results

• Four out of five models achieved accuracy above 90%.
• KNN performed poorly, mainly due to the high number of false negatives (57% sensitivity).
• Decision Tree and SVM excelled as the most efficient.
• Data processing and feature selection were key to the models' success.

7. Limitations and Future Directions

• Study limitations: Difficulty in finding standardized datasets (especially in Portuguese), few systems applied to the Brazilian context.
• Future directions: Test new algorithms (Naive Bayes, Boosting, K-means, Gradient Descent), apply other validation techniques, expand Portuguese datasets, include cross-validation (K-fold, Leave-one-out), and develop web applications for public use.

8. Conclusion

• Machine Learning has proven powerful for fake news detection and is crucial for protecting society from the impact of false information.
• Ongoing research is especially necessary in the Brazilian context.

9. Our Crew:

• 👩🏻‍🚀 Fabiana ⚡️ Campanari - Shoot me an email
• 👨🏽‍🚀 Pedro Barrenco

10. QR Code

11. References

Monteiro Bastos & Monteiro de Lima (2023). Fake News Detection Using Decision Tree, Support Vector Machine, and K-Nearest Neighbors Algorithms. Revista de Estudos Multidisciplinares XV Encontro Científico da UNDB.

💌 Let the data flow... Ping Me !

Contact and Support

• For notebook files, detailed tutorials, or enhanced visualizations, please reach out.
• Interested in Python notebooks simulating these dynamics or advanced Humanistic AI models? Just ask!

🛸๋ My Contacts Hub

────────────── 🔭⋆ ──────────────

➣➢➤ Back to Top

Copyright 2025 Mindful-AI-Assistants. Code released under the MIT license.