4.1 Feature Extractor Interface

Feature-extraction refers to the process of transforming raw-data into numerical features that can be processed whilst preserving the information in the original data set.

Feature Extraction is category driven whereby features are properties, attributes, or characteristics of a Record that are meaningfuly grouped together to form a category:

• Examples Features:

  • Accounts associated with the Record beloging to a certain list, i.e. wallet address that executed the transaction:
    • Feature: ACCOUNT_BLACKLISTED
  • Asset exposure to certain sources or activities such as Gambling, Money Laundering, etc.
    • Feature: ASSET_EXPOSURE_MONEYLAUNDERING
  • Volume of Assets transferred by a particular list of accounts
    • Feature: ACCOUNTLIST_ASSET_TRANSFER_VOLUME
  • Time of the day when the transaction was executed
    • Feature: TRANSACTION_TIME
  • If a certain internal call went against a specific Policy
    • Feature: POLICY_0X1A0B_VIOLATION

• Example Categories:

  • UNAPPROVED category that groups the following features:
    • POLICY_0X1A0B_VIOLATION
    • ACCOUNTLIST_ASSET_TRANSFER_VOLUME
  • APPROVED category that groups the following features:
    • ACCOUNT_BLACKLISTED
    • ASSET_EXPOSURE_MONYLAUNDERING
  • SUSPICIOUS category that groups the following features:
    • ACCOUNT_BLACKLISTED
    • ASSET_EXPOSURE_MONEYLAUNDERING

The Feature Extractor is responsible for extracting these features from the Record and delivering them in a structured format. It’s interface should be minimalistic and easy to implement.

Below is an example interface for the Feature Extractor:

package extractor

/// FeatureExtractorMetadata
/// provides metadata about the Feature Extractor
type FeatureExtractorMetadata interface {
	getName() string
	getDescription() string
	getVersion() string
	getAuthor() string
	getLicense() string
	getURL() string
	// getFeatureSchema returns
	// the serialized JSON schema for the Feature
	getFeatureSchema() ([]byte)
	getPluginList() ([]string)
}

type FeatureExtractor interface {
	FeatureExtractorMetadata
	// extractFeatures extracts
	// features from the given Record
	extractFeatures([]byte) ([]byte)
}

The extractFeatures function is the core method which processes a given Record object and returns data required for the classification process.