Country Classification

Correct geocoding is crucial for addressing people, businesses, properties, and other geographical points of interest to make smart business decisions, urban planning, and other governmental services. Yet, the lack of address data, such as the absence of the country name in the address, can negatively affect the accuracy of geocoding and cause trouble finding the exact location in the system.

This deep learning algorithm solves the issue as it can automatically predict the country for a certain incomplete street address. With this approach, the system will have the ability to identify the country before geocoding to increase address matching and eliminate possible errors and ambiguity.

The algorithm has been trained on address data obtained from OpenAddresses.io. It classifies addresses based on 18 different countries.

The prediction of the country may become an integral part of the pre-processing phase of a geocoding workflow. The address may then be fed to a geocoding service with added geographical context that could lead to enhanced search accuracy and increased efficiency of processing.

The algorithm may become useful in such fields as GIS, geospatial analytics, address intelligence, mapping, location-based services, and data cleansing. Incomplete addresses may be supplemented with a country name by using the model.