electricsheepafrica/Nigeria-conflict-events-1997-2025

Nigeria Conflict Events (ACLED) - ML Ready

Dataset Description

This is a cleaned and feature-engineered version of the Nigeria Conflict Events dataset from the Armed Conflict Location & Event Data Project (ACLED). The data has been specifically prepared for machine learning applications with advanced feature engineering, missing values handled, and temporal features created.

Cleaning and Feature Engineering Process

Applied Transformations:

1. Data Cleaning

• Missing values imputed (0 for numeric, 'Unknown' for categorical)
• Text columns standardized and trimmed
• Data types optimized for ML

2. Temporal Feature Engineering

• Date features extracted (year, month, quarter, days since start)
• Cyclical features for seasonality (sin/cos transformations)
• Lag features (previous month, previous year)
• Percentage changes and differences

3. Geographic Feature Engineering

• Administrative divisions encoded (Admin1, Admin2)
• Administrative-level aggregations (mean, std, max by region)

4. Event-Based Features

• Fatality-to-event ratios
• Log-transformed features for skewed distributions
• Binary indicators (has_events, has_fatalities, high_fatalities, high_events)
• Temporal changes and trends

5. Advanced Features

• Rolling statistics and aggregations
• Year-over-year comparisons
• Seasonal patterns captured through cyclical encoding

Dataset Summary

1. Political Violence Events (Cleaned)

• Original Shape: (268656, 10)
• Cleaned Shape: (268656, 43)
• Features: 43
• Time Period: 1997 - 2025

2. Civilian Targeting Events (Cleaned)

• Original Shape: (268656, 10)
• Cleaned Shape: (268656, 43)
• Features: 43
• Time Period: 1997 - 2025

3. Demonstration Events (Cleaned)

• Original Shape: (268656, 9)
• Cleaned Shape: (268656, 31)
• Features: 31
• Time Period: 1997 - 2025

Feature Categories

Temporal Features

• Date components (year, month, quarter)
• Cyclical encoding (sin/cos for seasonality)
• Days since start
• Lag features (previous month/year)
• Changes and percentage changes

Geographic Features

• Encoded administrative divisions
• Regional statistics and aggregations
• Location-based identifiers

Event-Based Features

• Raw event and fatality counts
• Log-transformed versions
• Ratios and derived metrics
• Binary indicators for thresholds

Advanced Features

• Rolling statistics
• Year-over-year comparisons
• Seasonal patterns
• Trend indicators

Files

• political_violence_cleaned.parquet: Cleaned and feature-engineered political violence data
• civilian_targeting_cleaned.parquet: Cleaned and feature-engineered civilian targeting data
• demonstrations_cleaned.parquet: Cleaned and feature-engineered demonstrations data
• data_dictionary.json: Detailed information about features and encoding
• metadata.csv: Original metadata file

Usage for ML

from datasets import load_dataset
import pandas as pd

# Load the cleaned dataset
dataset = load_dataset("electricsheepafrica/nigeria-conflict-events-ml-ready")

# Access cleaned data
political_violence = pd.read_parquet("political_violence_cleaned.parquet")

# Example: Prepare features for time series forecasting
features = [col for col in political_violence.columns if col not in ['Events', 'Fatalities', 'date']]
X = political_violence[features]
y_events = political_violence['Events']
y_fatalities = political_violence['Fatalities']

# Ready for ML models (scikit-learn, TensorFlow, PyTorch)

ML Applications

This cleaned dataset is ideal for:

Time Series Forecasting

• Event count prediction
• Fatality forecasting
• Conflict intensity modeling

Classification Tasks

• Event severity classification
• Hotspot identification
• Risk level categorization

Anomaly Detection

• Unusual event pattern detection
• Outlier identification
• Early warning systems

Spatial Analysis

• Geographic hotspot mapping
• Regional conflict patterns
• Administrative level analysis

Data Quality

• Missing Values: 0 (all imputed)
• Duplicate Rows: Removed
• Data Types: Optimized for ML
• Feature Scaling: Log transforms applied for skewed features
• Encoding: Categorical variables properly encoded

License

This dataset is subject to ACLED's data use terms and conditions. Please review ACLED's website for specific licensing information.

Citation

If you use this dataset, please cite the original source:

Armed Conflict Location & Event Data Project (ACLED). (2025). 
Nigeria Conflict Events. 
Retrieved from https://acleddata.com/

Processing Notes

• All features are numeric and ready for ML algorithms
• Temporal dependencies preserved through lag features
• Geographic relationships captured through encoding
• Seasonal patterns encoded through cyclical features
• Data is ready for direct use in any ML framework