liveBook · Manning

Feature vector

As explained in Fight Fraud with Machine Learning

A feature vector is an ordered collection of feature values representing one instance for a machine-learning model.

A feature vector is a fixed-length collection of values representing one transaction for a machine-learning model. In a Bitcoin transaction graph, each transaction node has 165 features describing transaction metadata and aggregated information from neighboring transactions. section 9.1.2

Inspecting feature vectors reveals the model input and its relationship to graph structure:

  • Check the dataset feature count before modeling; the Bitcoin dataset reports 165 features per transaction. section 9.1.2
  • Separate node features from edges: the vector describes an individual transaction, while directed edges represent payment flows between transactions. section 9.1.2
  • Use feature vectors alone as input to a fully connected model, or combine them with neighborhood connections for graph learning. section 9.1.4

Bitcoin transaction features

The first 94 features capture local information such as timestamps, input and output counts, transaction fees, and average BTC amounts. The remaining features aggregate one-hop neighbor statistics, including maximum, minimum, standard deviation, and correlation values. section 9.1.2

How Fight Fraud with Machine Learning covers it

How deep it goes
Defines a 165-value transaction representation and its metadata sources, then contrasts feature-only modeling with neighborhood-based graph learning.
Examples
Uses Bitcoin transaction metadata and one-hop neighbor statistics, including fees, input/output counts, averages, extrema, deviations, and correlations.
What it assumes
Assumes familiarity with transactions, graph nodes and edges, and binary labels such as licit and illicit.
What you can do after
Apply it via feature inspection and feature-only modeling to classify illicit transactions before incorporating graph neighborhoods.
How it connects
Connects node features with transactions, directed graphs, graph learning, multilayer perceptrons, and binary classification.

Also explained in

Compare how 5 books explain feature vector →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Fight Fraud with Machine Learning. The section links point to the parts of the book it draws on.