README.md

---
license: mit
task_categories:
- tabular-classification
language:
- en
tags:
- social-media
- spam-detection
- facebook
- cybersecurity
- machine-learning
- binary-classification
- fraud-detection
size_categories:
- n<1K
---

# Facebook Spam Detection Dataset

## Dataset Summary

This dataset contains **600 Facebook profiles** with behavioral and activity features designed for **spam detection** in social media. The dataset enables binary classification to distinguish between spam accounts (Label=1) and legitimate accounts (Label=0), providing insights into spammer behavior patterns on Facebook.

## Dataset Details

- **Total Samples**: 600 profiles
- **Classes**: Binary (0 = Legitimate, 1 = Spam)
- **Class Distribution**: Imbalanced (17.2% spam, 82.8% legitimate)
- **Features**: 14 behavioral characteristics + 1 target label
- **Format**: CSV file

## Features Description

| Feature | Type | Description | Range |
|---------|------|-------------|-------|
| `profile id` | Integer | Unique profile identifier | 1-600 |
| `#friends` | Integer | Number of friends | 4-5,554 |
| `#following` | Integer | Number of accounts being followed | 1-5,312 |
| `#community` | Integer | Number of communities/groups joined | 12-1,789 |
| `age` | Integer | Account age (likely in days) | 125-2,697 |
| `#postshared` | Integer | Total number of posts shared | 76-3,896 |
| `#urlshared` | Integer | Number of URLs shared in posts | 11-2,956 |
| `#photos/videos` | Integer | Number of photos/videos posted | 65-3,891 |
| `fpurls` | Float | Frequency/proportion of URLs in posts | 0.01-1.09 |
| `fpphotos/videos` | Float | Frequency/proportion of media content | 0.0-2.74 |
| `avgcomment/post` | Float | Average comments per post | 0.0-665 |
| `likes/post` | Float | Average likes per post | 0.1-2.8 |
| `tags/post` | Integer | Tags used in posts | 10-99 |
| `#tags/post` | Integer | Number of tags per post | 1-32 |
| `Label` | Integer | **Target variable** - Spam (1) or Legitimate (0) | 0-1 |

## Key Statistics

- **Network Size**: Average 1,066 friends and 1,069 following
- **Community Engagement**: Average 208 communities joined
- **Account Maturity**: Average age of 1,215 days (~3.3 years)
- **Content Activity**: 
  - Average 1,158 posts shared
  - Average 370 URLs shared
  - Average 1,121 photos/videos posted
- **Engagement Metrics**:
  - Average 1.6 comments per post
  - Average 0.88 likes per post
  - Average 16 tags per post

## Class Imbalance

⚠️ **Important**: This dataset is imbalanced:
- **Legitimate accounts**: 497 samples (82.8%)
- **Spam accounts**: 103 samples (17.2%)

Consider using techniques like SMOTE, class weighting, or balanced sampling for training.

## Use Cases

This dataset is ideal for:

- **Spam Detection**: Build classifiers to identify Facebook spam accounts
- **Behavioral Analysis**: Study differences between spam and legitimate user behavior
- **Anomaly Detection**: Develop unsupervised methods for suspicious activity detection
- **Social Media Security**: Research automated content moderation systems
- **Imbalanced Learning**: Practice techniques for handling skewed datasets

## Quick Start

```python
import pandas as pd
from sklearn.model_selection import train_test_split
from sklearn.ensemble import RandomForestClassifier
from sklearn.metrics import classification_report, confusion_matrix
from imblearn.over_sampling import SMOTE

# Load dataset
df = pd.read_csv('Facebook Spam Dataset.csv')

# Prepare features and target
X = df.drop(['Label', 'profile id'], axis=1)
y = df['Label']

# Handle class imbalance with SMOTE
smote = SMOTE(random_state=42)
X_resampled, y_resampled = smote.fit_resample(X, y)

# Split data
X_train, X_test, y_train, y_test = train_test_split(
    X_resampled, y_resampled, test_size=0.2, random_state=42, stratify=y_resampled
)

# Train model
model = RandomForestClassifier(
    n_estimators=100, 
    class_weight='balanced',
    random_state=42
)
model.fit(X_train, y_train)

# Evaluate
y_pred = model.predict(X_test)
print("Classification Report:")
print(classification_report(y_test, y_pred))
```

## Suggested Approaches

### Traditional ML
- **Random Forest**: Handles mixed data types well
- **Gradient Boosting**: XGBoost, LightGBM for performance
- **SVM**: With RBF kernel for non-linear patterns
- **Logistic Regression**: With proper feature scaling

### Handling Imbalance
- **Sampling**: SMOTE, ADASYN for oversampling
- **Cost-sensitive**: Class weights in algorithms
- **Ensemble**: Balanced bagging, EasyEnsemble
- **Metrics**: Focus on F1-score, AUC-ROC, precision/recall

### Feature Engineering
- **Ratios**: Create engagement ratios (likes/posts, comments/posts)
- **Behavioral**: URL sharing patterns, media content ratios
- **Network**: Friend-to-following ratios, community participation
- **Temporal**: Account age interactions with activity levels

## Model Evaluation Tips

Given the class imbalance, prioritize these metrics:
- **F1-Score**: Harmonic mean of precision and recall
- **AUC-ROC**: Area under the ROC curve
- **Precision/Recall**: Especially for spam class (minority)
- **Confusion Matrix**: To understand false positives/negatives

## Data Quality

- ✅ **Complete Data**: No missing values
- ⚠️ **Class Imbalance**: 82.8% legitimate vs 17.2% spam
- ✅ **Feature Variety**: Network, content, and engagement metrics
- ✅ **Realistic Ranges**: All features show plausible Facebook activity patterns

## Research Opportunities

1. **Behavioral Patterns**: What distinguishes spam from legitimate user behavior?
2. **Feature Importance**: Which metrics are most predictive of spam accounts?
3. **Temporal Analysis**: How does account age correlate with spam likelihood?
4. **Network Effects**: Do spam accounts show distinct networking patterns?
5. **Content Analysis**: How do URL sharing and media patterns differ?

## Potential Applications

- **Social Media Platforms**: Automated spam account detection
- **Content Moderation**: Flagging suspicious posting patterns
- **User Safety**: Protecting users from spam and malicious content
- **Research**: Understanding social media abuse patterns
- **Security Systems**: Real-time threat detection algorithms

## Citation

```bibtex
@dataset{facebook_spam_detection_2024,
  title={Facebook Spam Detection Dataset},
  year={2025},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/nahiar/facebook-spam-detection}
}
```

## Notes

- The `age` feature appears to be in days rather than years
- Some ratio features (like `fpurls`, `fpphotos/videos`) may exceed 1.0, indicating normalized metrics
- Consider feature scaling for distance-based algorithms
- The dataset reflects Facebook's ecosystem and user behavior patterns