README for Part 2 solution

Soldier Defection Risk Synthetic Dataset Generation (Kindly Navigate to the Part 2 folder)

This repository contains a Python script that generates a synthetic dataset of 10,000 soldiers to model and predict the likelihood of military defection based on various individual and external factors. The dataset is designed to simulate real-world scenarios and is useful for research, analysis, and modeling purposes.

---

Table of Contents

• Introduction
• Features
  • Individual-Level Factors
  • External-Level Factors
• Dataset Generation Process
  • 1. Import Libraries
  • 2. Initialize the Dataset
  • 3. Generate Features
    • Security Clearance Level
    • Morale Score
    • Family Military History
    • Punishment Policy
    • Regime Type
    • Trust in Leadership Score
    • Military Structure
    • Defector Capture Rate
    • Promotion Fairness Score
    • Communication Quality Score
    • Opportunity Cost
    • Has Enemy Connections
    • Time of Measurement
• Defection Risk Calculation
  • Feature Weights
  • Normalization
  • Defection Risk Score
  • Threshold Determination
  • Final Classification
• Usage
• Conclusion
• Next Steps

---

Introduction

Military defection is a complex phenomenon influenced by a multitude of factors at both the individual and external levels. Understanding these factors is crucial for predicting defection risk and implementing strategies to maintain military cohesion. This project simulates a dataset that captures these factors, allowing for analysis and modeling of defection behaviors.

---

Features

The dataset includes a variety of features that are divided into individual-level and external-level factors. Each feature is carefully generated to reflect realistic distributions and relationships with the likelihood of defection.

Individual-Level Factors

1. Opportunity Cost (opportunity_cost_norm): The potential benefits a soldier foregoes by remaining in service. Higher values increase defection risk.

2. Family Military History (family_military_history_bin): Indicates if a soldier has family members who served in the military. Contrary to intuition, in this model, it positively correlates with defection risk.

3. Security Clearance Level (security_clearance_level_num): Access to sensitive information. Higher clearance levels increase defection risk due to the value of the information.

4. Has Enemy Connections (has_enemy_connections_bin): Indicates connections with enemy forces. Presence of such connections increases defection risk.

5. Morale Score (morale_score_norm): Overall satisfaction and morale. Higher morale decreases defection risk.

6. Trust in Leadership Score (trust_in_leadership_score_norm): Trust in military leadership. Higher trust decreases defection risk.

External-Level Factors

1. Punishment Policy (punishment_policy_num): Perceived strictness of anti-defection policies. Lenient policies increase defection risk.

2. Regime Type (regime_type_num): Type of political regime. Personalist regimes have higher defection risk.

3. Military Structure (military_structure_num): Level of institutionalization. Institutionalized structures have higher defection risk due to formal systems.

4. Defector Capture Rate (defector_capture_rate_norm): Rate at which defectors are captured. Higher rates decrease defection risk.

5. Promotion Fairness Score (promotion_fairness_score_norm): Fairness of promotion systems. Fair systems decrease defection risk.

6. Communication Quality Score (communication_quality_score_norm): Quality of communication within the military. Better communication decreases defection risk.

---

Dataset Generation Process

The dataset is generated through a series of steps, each responsible for creating and processing different features.

1. Import Libraries

import numpy as np
import pandas as pd

# Set random seed for reproducibility
np.random.seed(0)

2. Initialize the Dataset

# Number of soldiers
n = 10000

# Initialize empty DataFrame
df = pd.DataFrame()

3. Generate Features

Security Clearance Level

• Description: Represents the security clearance of each soldier.
• Values: 'low' (60%), 'medium' (30%), 'high' (10%).

df['security_clearance_level'] = np.random.choice(
    ['low', 'medium', 'high'], size=n, p=[0.6, 0.3, 0.1]
)
# Map to numerical values
security_clearance_mapping = {'low': 1, 'medium': 2, 'high': 3}
df['security_clearance_level_num'] = df['security_clearance_level'].map(security_clearance_mapping)

Morale Score

• Description: Represents the morale level of each soldier.
• Values: Random float between 0 and 100.

df['morale_score'] = np.random.uniform(0, 100, size=n)

Family Military History

• Description: Indicates if a soldier has family members who served in the military.
• Values: Binary (0 or 1), with a 10% chance of being 1.

df['family_military_history_bin'] = np.random.binomial(1, 0.1, size=n)

Punishment Policy

• Description: Perceived strictness of anti-defection policies.
• Values: 'strict' (80%), 'lenient' (20%).

df['punishment_policy'] = np.random.choice(
    ['strict', 'lenient'], size=n, p=[0.8, 0.2]
)
# Encode to numerical values
df['punishment_policy_num'] = df['punishment_policy'].map({'strict': 0, 'lenient': 1})

Regime Type

• Description: Type of political regime.
• Values: 'personalist'.

df['regime_type'] = 'personalist'
# Encode to numerical values
df['regime_type_num'] = df['regime_type'].map({'personalist': 1, 'party-based': 0})

Trust in Leadership Score

• Description: Level of trust in military leadership.
• Values: Random float between 0 and 100.

df['trust_in_leadership_score'] = np.random.uniform(0, 100, size=n)

Military Structure

• Description: Level of institutionalization in the military.
• Values: 'patrimonial'.

df['military_structure'] = 'patrimonial'
# Encode to numerical values
df['military_structure_num'] = df['military_structure'].map({'patrimonial': 0, 'institutionalized': 1})

Defector Capture Rate

• Description: Rate at which defectors are captured.
• Values: Constant at 0.8.

df['defector_capture_rate'] = 0.8

Promotion Fairness Score

• Description: Fairness of the promotion system.
• Values: 80% between 0 and 40, 20% between 40 and 100.

df['promotion_fairness_score'] = np.where(
    np.random.rand(n) < 0.8,
    np.random.uniform(0, 40, size=n),
    np.random.uniform(40, 100, size=n)
)

Communication Quality Score

• Description: Quality of communication within the military.
• Values: Random float between 0 and 100.

df['communication_quality_score'] = np.random.uniform(0, 100, size=n)

Opportunity Cost

• Description: Potential benefits a soldier foregoes by remaining in service.
• Values: Random float between 0 and 5000, adjusted based on security clearance.

df['opportunity_cost'] = np.random.uniform(0, 5000, size=n)
# Adjust based on security clearance
df['opportunity_cost'] = df.apply(
    lambda row: row['opportunity_cost'] if row['security_clearance_level'] == 'low' else
                row['opportunity_cost'] * 0.5 if row['security_clearance_level'] == 'medium' else
                row['opportunity_cost'] * 0.2,
    axis=1
)
# Normalize
df['opportunity_cost_norm'] = (
    df['opportunity_cost'] - df['opportunity_cost'].min()
) / (df['opportunity_cost'].max() - df['opportunity_cost'].min())

Has Enemy Connections

• Description: Indicates if a soldier has connections with enemy forces.
• Values: 10% 'Yes', 90% 'No', with 95% of values set to NaN to reflect missing data.

df['has_enemy_connections'] = np.where(
    np.random.binomial(1, 0.1, size=n) == 1, 'Yes', 'No'
)
# Set 95% to NaN
missing_mask = np.random.rand(n) < 0.95
df.loc[missing_mask, 'has_enemy_connections'] = np.nan
# Encode to binary
df['has_enemy_connections_bin'] = df['has_enemy_connections'].map({'Yes': 1, 'No': 0}).fillna(0)

---

Defection Risk Calculation

Feature Weights

Weights are assigned to each feature based on their positive or negative relationship with defection risk.

• Positive Relationship (increase defection risk):

  Feature	Weight
  opportunity_cost_norm	+0.4
  security_clearance_level_num	+0.3
  has_enemy_connections_bin	+0.5
  punishment_policy_num	+0.2
  regime_type_num	+0.3
  military_structure_num	+0.2

• Negative Relationship (decrease defection risk):

  Feature	Weight
  family_military_history_bin	-0.2
  morale_score_norm	-0.3
  trust_in_leadership_score_norm	-0.3
  promotion_fairness_score_norm	-0.2
  communication_quality_score_norm	-0.1
  defector_capture_rate_norm	-0.4

Normalization

Features are normalized to ensure they are on the same scale (0 to 1).

# Normalize negative relationship features
df['morale_score_norm'] = df['morale_score'] / 100
df['trust_in_leadership_score_norm'] = df['trust_in_leadership_score'] / 100
df['promotion_fairness_score_norm'] = df['promotion_fairness_score'] / 100
df['communication_quality_score_norm'] = df['communication_quality_score'] / 100
df['defector_capture_rate_norm'] = df['defector_capture_rate']  # Already between 0 and 1

# Normalize security clearance level
df['security_clearance_level_norm'] = df['security_clearance_level_num'] / df['security_clearance_level_num'].max()

Defection Risk Score

The defection risk score is calculated by summing the weighted contributions of each feature.

df['defection_risk_score'] = (
    weights['opportunity_cost_norm'] * df['opportunity_cost_norm'] +
    weights['family_military_history_bin'] * df['family_military_history_bin'] +
    weights['security_clearance_level_num'] * df['security_clearance_level_norm'] +
    weights['has_enemy_connections_bin'] * df['has_enemy_connections_bin'] +
    weights['morale_score_norm'] * df['morale_score_norm'] +
    weights['trust_in_leadership_score_norm'] * df['trust_in_leadership_score_norm'] +
    weights['promotion_fairness_score_norm'] * df['promotion_fairness_score_norm'] +
    weights['communication_quality_score_norm'] * df['communication_quality_score_norm'] +
    weights['defector_capture_rate_norm'] * df['defector_capture_rate_norm'] +
    weights['punishment_policy_num'] * df['punishment_policy_num'] +
    weights['regime_type_num'] * df['regime_type_num'] +
    weights['military_structure_num'] * df['military_structure_num']
)

Threshold Determination

A threshold is set based on the median defection risk score.

threshold = df['defection_risk_score'].median()

Final Classification

Soldiers are classified as 'yes' (will defect) or 'no' (will not defect) based on whether their defection risk score exceeds the threshold.

df['will_defect'] = np.where(df['defection_risk_score'] > threshold, 'yes', 'no')

---

Usage

1. Clone the Repository:

   git clone https://github.com/yourusername/defection-risk-dataset.git
   cd defection-risk-dataset

2. Install Dependencies:

   Ensure you have Python 3.x installed along with numpy and pandas.

   pip install numpy pandas

3. Run the Script:

   python generate_dataset.py

4. Explore the Dataset:

   The script will generate soldier_defection_dataset.csv containing the synthetic data.

---

Conclusion

This project provides a comprehensive synthetic dataset that models the complex factors influencing military defection. By assigning weights based on domain knowledge and normalizing features, we create a realistic simulation useful for analytical and predictive purposes.

---

Next Steps

• Model Training: Use the dataset to train machine learning models to predict defection.

• Validation: Validate the synthetic data and model predictions against real-world data if available.

• Feature Refinement: Adjust feature weights and distributions based on additional research or data.

• Temporal Analysis: Incorporate the time_of_measurement feature to study how defection risk changes over time.

• Scenario Simulation: Modify constant features like defector_capture_rate and regime_type to simulate different scenarios and their impact on defection risk.

---

Contact

For questions or suggestions, please contact official.tanmay1306@gmail.com.

---

Disclaimer: This dataset is synthetic and created for educational and research purposes. It does not represent real individuals or events.