No ArcGIS, No QGIS—Just Python and a Problem Worth Solving

Rahmah Abubakar — Sun, 10 Aug 2025 15:33:27 +0000

This was 2024 project I did during my internship with HNG TECH focused on data cleansing and anomaly detection in electoral datasets. I recently revisited it to document the methodology and insights more clearly, as part of building my professional portfolio and showcasing real-world analytical impact.

Section 1: Why This Matters**

Electoral credibility is foundational to democracy. In this project, I used Python to analyze polling unit-level data from Zamfara State, identifying statistical anomalies that could signal irregularities, data entry errors, or procedural lapses. This is a lightweight, reproducible approach to election auditing—no GIS tools required.

Section 2: Methodology Overview**

📥 Data Source

CSV file with polling unit-level results
Fields: Latitude, Longitude, Registered Voters, Accredited Voters, Votes per Party, PU Metadata.
Excel was used to manually clean and organize the geospatial coordinates (longitude and latitude)

Outlier Detection

Calculated outlier scores using z-scores and domain heuristics
Ranked polling units by anomaly severity per party

Section 2: Key Findings

PU 108 & PU 104 (Birnin Magaji): High outlier scores for APC and PDP
PU 321 & PU 124 (Tsafe): Negative transcription counts flagged
Geospatial Clustering: Anomalies concentrated in specific LGAs
Cross-Party Deviations: Irregularities not isolated to one party

Section 4: Technical Highlights

No GIS tools—just Python (Pandas, Matplotlib, NumPy)
Google Colab for cloud execution
Scalable framework for other states or elections

Section 5: Recommendations

Audit flagged polling units
Fix metadata gaps and transcription errors
Integrate anomaly detection into official result verification

Closing Thoughts

This project taught me how to apply statistical thinking to real-world problems, especially in politically sensitive domains. It’s a reminder that data science isn’t

For full analysis—including code, tables, and transcription breakdowns—see the complete notebook on my LinkedIn page(https://www.linkedin.com/posts/rahmah-abubakar-243058288_analysis-process-activity-7253521166152163328-zh2n?utm_source=share&utm_medium=member_android&rcm=ACoAAEXGTFMBU8WaWDL8Z4Wl7HzBBVCiQx_eYO4)._

DEV Community: Rahmah Abubakar

No ArcGIS, No QGIS—Just Python and a Problem Worth Solving