Shoumique Hasan

DRIVER

Thirty years of road incident data. One migration. Zero data loss.

BJIT, World Bank / JICA, 2024 to 2025

Legacy state

MS-DOS

thirty years, 1990 to 2020

Legacy format, Hard to query and analyze

Modern platform

0%

data loss

Automated transformation, Geospatial and ML-enabled

The problem

The Dhaka Metropolitan Police held thirty years of road incident data, from 1990 to 2020, locked inside an MS-DOS-era software system. That history held the patterns needed to understand where and why incidents happen. In its legacy form, it was effectively unreachable.

Legacy system

The source was an MS-DOS software system. Three decades of records sat in a legacy format that couldn’t be queried, analyzed or extended. Every passing year added more data, and more risk that it would stay stuck where it was.

The challenge

Copying the data wasn’t the hard part. The hard part was transforming decades of legacy records into the clean, relational, geospatial shape the new platform required, without losing a single record in the process. The acceptance bar was zero data loss.

Migration approach

The transformation and migration were automated. Rather than hand-carrying three decades of records, the process read the legacy data, transformed it into the new platform’s model, and wrote it into a Django REST / GeoDjango system running on AWS.

MS-DOS System 1990 to 2020 source Extract read legacy records Transform normalize and validate New platform Django REST, GeoDjango Predictive models danger and red zones automated migration 0% data loss geospatial (GeoDjango) ML forecasting

Scroll to see the full diagram.

  • Automated migration
  • Zero data loss
  • Geospatial platform
  • ML forecasting

Zero data loss

Note

Zero data loss wasn’t a target to aim for: it was the acceptance criterion. Records were reconciled against the source so nothing fell through the cracks during transformation.

Every record had to make it across. The new system had to be a faithful superset of the old one: three decades of history, intact and now queryable.

Predictive analytics

Note

The same thirty years of data that had to migrate became training signal. Historical incidents were turned into forward-looking forecasts of danger and red-zone areas.

Machine learning models were developed and integrated into the system to analyze incident data and predict danger areas, turning a static archive into an analytical asset for the police.

Results

The migration completed without disruption to the new system, and the work was rewarded with a performance bonus. Thirty years of road incident data moved from a legacy MS-DOS system into a modern, geospatially aware, ML-enabled platform.