COVID-19 Contact Tracing — Anonymized Mobile Data

Authored a detailed research piece on using anonymized mobile phone data for real-time contact tracing during the COVID-19 pandemic, published on AI Hub in June 2020. The work explored how telecom operators’ location datasets — combined with AI algorithms — could model individual and social interactions to support disease containment, while rigorously addressing the ethical and privacy dimensions of such an approach.

1. Model population movement using mobile location data

Drawing on prior research from the Haiti earthquake and malaria outbreaks in Tanzania, the analysis demonstrated how SIM card position data can accurately track large-scale population migration. Mobile-based models outperformed traditional gravity models in predicting outbreak spatial spread (AUC 0.79 vs 0.66–0.74).

# Example: Simplified mobility-based infection risk scorer
import pandas as pd
import numpy as np

def score_infection_risk(mobility_df: pd.DataFrame, case_counts: pd.DataFrame) -> pd.DataFrame:
    merged = mobility_df.merge(case_counts, on="region_id")
    merged["risk_score"] = (
        merged["avg_daily_visitors"] * merged["active_cases"] /
        merged["population"]
    )
    merged["risk_tier"] = pd.cut(
        merged["risk_score"],
        bins=[0, 0.01, 0.05, float("inf")],
        labels=["low", "medium", "high"]
    )
    return merged.sort_values("risk_score", ascending=False)

2. Analyze R0 dynamics and intervention effects

Examined how lockdowns and social distancing measures affected the reproduction number (R0) in real time. In Wuhan, R0 values dropped well below the epidemic threshold during lockdown — a pattern observable through mobile activity data.

3. Address data privacy and ethical constraints

A central theme of the research was balancing public health utility against individual privacy. The piece highlighted GDPR-style protections as a necessary global standard and advocated for anonymization-first approaches rather than individual tracking.

# Example: Anonymization check — ensure no single-user cells survive aggregation
def validate_anonymization(df: pd.DataFrame, geo_col="cell_id", min_users=5) -> dict:
    counts = df.groupby(geo_col).size().reset_index(name="user_count")
    violations = counts[counts["user_count"] < min_users]
    return {
        "total_cells": len(counts),
        "violations": len(violations),
        "safe": len(violations) == 0
    }

4. Identify future scope and limitations

Proposed concrete improvements for next-generation contact tracing: incorporating behavior change data, profession-based contact matrices, and environmental factors like humidity and temperature — while acknowledging coverage gaps in regions with low smartphone penetration.