ActivityTypeClassifier

public class ActivityTypeClassifier: MLClassifier

Activity Type Classifiers are Machine Learning Classifiers. Use an Activity Type Classifier to determine the ActivityTypeName of a LocomotionSample.

Supported Activity Types

Base Types

stationary, transport, walking, running, cycling

Base types match one-to-one with Core Motion activity types, with the exception of Core Motion’s automotive being renamed to transport in LocoKit.

Extended Types

car, train, bus, motorcycle, airplane, boat

Extended types are a subset of the base transport type, allowing for more specific classification when enough local data is available.

Region Specific Classifiers

LocoKit provides geographical region specific machine learning data, with each classifier containing the data for a specific region.

This allows for detecting activity types based on region specific characteristics, with much higher accuracy than iOS’s built in Core Motion types detection. It also makes it possible to detect a greater number of activity types, for example distinguishing between travel by car or train.

LocoKit’s data regions are roughly 100 kilometres by 100 kilometres squared (0.1 by 0.1 degrees), or about the size of a small town, or a single neighbourhood in a larger city.

Larger cities might encompass anywhere from four to ten or more classifier regions, thus allowing the classifers to accurately detect activity type differences within different areas of a single city.

Determining Regional Coverage

• LocoKit transport coverage maps
• LocoKit cycling coverage maps

Stationary, Transport, Walking, Running

The base activity types of stationary, transport, walking, and running do not significantly differ by geographical region, thus should achieve high detection accuracy everywhere in the world, regardless of local data availability.

These types can be considered to have global coverage.

Cycling

Cycling has enough regional variance in locomotive characteristics that detection accuracy can range from excellent to average depending on the availability of local model data.

If very high accuracy cycling detection is important to your application, you should check the cycling coverage map for the regions you require. However if cycling detection is not a core function of your app, then the results from even low coverage classifiers should achieve adequate accuracy, and will certainly exceed the accuracy of Core Motion’s detection.

Car, Train, Bus, Motorcycle, Airplane, Boat

While the base transport type can be detected anywhere in the world with high accuracy, determining the specific mode of transport requires local knowledge. If knowing the specific mode of transport is important to your application, you should check the coverage maps for your required regions.