paint-brush
Five Real Machine Learning Use Cases in Cryptocurrenciesby@jrodthoughts
2,250 reads
2,250 reads

Five Real Machine Learning Use Cases in Cryptocurrencies

by Jesus RodriguezSeptember 25th, 2019
Read on Terminal Reader
Read this story w/o Javascript
tldt arrow

Too Long; Didn't Read

Machine Learning is an algorithm that can learn from data without relying on rules-based programming. Machine learning can enable models that learn the behavior of well-known crypto exchanges and identify new ones. Data rich nature of crypto-assets creates some new scenarios that can be tackled using machine learning. Machine Learning can identify significant patterns in a given crypto-asset given a unique canvas of a given data set. Using machine learning to identify groups of groups of investors and groups of token holders that qualify their behavior, unvised models can be used to forecast the forecast.

Company Mentioned

Mention Thumbnail

Coins Mentioned

Mention Thumbnail
Mention Thumbnail
featured image - Five Real Machine Learning Use Cases in Cryptocurrencies
Jesus Rodriguez HackerNoon profile picture

We hear this all the time: a new analytics platform or study that uses machine learning to analyze crypto-assets. However, when we dig a bit deeper, instead of cutting edge machine learning we find simple statistics or basic algebra glorified as a sophisticated analysis.

You would think that the use of machine learning methods in crypto-assets would be ubiquitous. After all, cryptocurrencies operate in a largely digital environment governed by public ledgers and we are living in the golden era of machine learning and artificial intelligence(AI). However, the use cases for applying machine learning in cryptocurrencies remain a novelty. At IntoTheBlock, we spent a lot of time applying different machine learning methods to specific problems in the crypto space.

Today, I would like to explore some practical use cases in which machine learning can deliver tangible value in the crypto-asset space.

Machine Learning vs. Statistics

Labeling something as machine learning or AI is sexy these days and the crypto analytics market is not exempt of that hype. In the case of the crypto-asset market, most of the analytics we’ve seen are relatively simple statistics glorified as machine learning techniques for marketing purposes. No harm on that; after all, the line between statistics and machine learning is blurry in some areas and some experts even refer to machine learning as “glorified statistics”.

To understand what methods can be classified as machine learning and which ones shouldn’t, it is key to make the distinction between machine learning and statistical techniques.

There are plenty of elements that can be cited to explain the differences between machine learning and statistics. Conceptually, both disciplines are focused on the analysis, interpretation and organization of patterns in datasets. However, statistics achieves its goal by deriving inferences in the form of mathematical equations while machine learning creates models that have the ability to learn beyond the programmed code. In a nutshell:

Machine Learning is an algorithm that can learn from data without relying on rules-based programming.Statistical modeling is a formalization of relationships between variables in the data in the form of mathematical equations.

We can also think about it these way: “Statistics draws population inferences from a sample, and machine learning finds generalizable predictive patterns.”

In the context of crypto-assets there are plenty of opportunities for applying both statistics and machine learning techniques. However, the use cases for the latter might not be that obvious at first glance. Let’s explore some of the most common scenarios for applying machine learning methods to crypto-assets.

Five Machine Learning Scenarios for Crypto-Assets

When we think about machine learning and cryptocurrencies, there are some use cases such as price prediction or sentiment analysis that immediately come to mind. However, those use cases are not specific to crypto-assets and the machine learning techniques applied are not different from those in public equities or other financial markets. However, the data rich nature of crypto-assets creates some new scenarios that can be tackled using machine learning.

1) Bad Actor Detection

· Problem Description: Is a transaction fraudulent or not? Identifying fraudulent behavior or financial crimes taking place with cryptocurrencies is a key scenario for financial authorities.

· Why Machine Learning?: The traditional way to tackle this problem will be to train models in well-known fraudulent patterns using different factors so that they can learn and generalize that knowledge and identify similar fraudulent activities.

· Machine Learning Techniques: Clustering, Classification, Decision Trees.

2)Exchange Classification

· Problem Description: Is this address an exchange or an individual wallet? Identifying addresses belonging to centralized exchanges as well as its corresponding transactions can help to predict all sorts of behaviors in crypto-assets.

· Why Machine Learning?: There are no large labeled datasets of cryptocurrency exchanges. Machine learning can enable models that learn the behavior of well-known crypto exchanges and identify new ones.

· Machine Learning Techniques: Clustering, Classification, Decision Tree.

3) Wallet Behavior Analysis

· Problem Description: How does this investor or group of investors buys or sell a specific crypto-asset? Understanding the patterns and behavior of individual investors is one of the unique benefits of crypto-assets. Individual investors or similar groups of investors tend to be more predictable than a crypto-asset as a whole.

· Why Machine Learning?: Using machine learning we can identify relevant groups of investors and identify patterns that qualify their behavior. Specifically, unsupervised models can identify significant patterns in a given group of token holders as well as unique factors that describe their behavior. Based on that predictive models can be used to forecast the behavior of individual investors.

· Machine Learning Techniques: Recurrent neural networks, convolutional neural networks.

4) On-Chain Power Factors

· Problem Description: Are these factors predictors of the behavior of a given crypto-asset? Blockchain datasets offer a unique canvas to identify unique factors that result influential in the behavior of a given crypto-asset. By new factors, I am referring to characteristics such as hash rate, or minig rewards distributions that are unique to crypto-assets.

· Why Machine Learning?: Using machine learning we can identify relevant groups of investors and identify patterns that qualify their behavior.

· Machine Learning Techniques: Linear regressions, decision trees, recurrent neural networks.

5) Fund Flow Analysis

· Problem Description: Are money transfers between exchanges signaling a position in a given crypto-asset? Are whales accumulating or dumping? Are miners distributing rewards in an unusual way? Blockchain datasets offers a unique opportunity to analyze the flow of funds between known entities.

· Why Machine Learning?: Machine learning can help us understand flow of funds patterns between known actors and try to predict potential impacts in price. For instance, we can detect correlations between regular transfers between two exchanges with large orders being placed.

· Machine Learning Techniques: Long-Short-Term-Memory networks, recurrent neural networks.

As you can see, there are plenty of unique scenarios for applying machine learning techniques to understanding and ultimately predicting the behavior of crypto-assets. Blockchain datasets certainly offer a blank canvas for machine learning experimentation in crypto-assets and the insights revealed by these type of techniques are nothing short of remarkable.

More than in any other asset class in the past, machine learning and data science are called to play a prominent role in the evolution of crypto-assets.

(Disclaimer: The author is the CTO at IntoTheBlock)