How Token Metrics Uses AI
Guide on How Token Metrics uses Machine Learning
Written by Saikiran Sagiraju
Updated over a week ago

At Token Metrics, our machine learning models take inspiration from genetics and, more specifically, evolutionary natural selection.

To better understand how it works, consider this example:

You want to breed giraffes with long necks.

You start by taking a random sample of 100 giraffes, score them and rank them based on how long their necks are.

You then take the best two giraffes from your sample and let them mate to create a new generation of 100 giraffes.

Repeat this process for 500 generations, and after that, you will most likely have giraffes with very long necks.

That’s what we do with our models for crypto tokens!

Let’s get more specific:

Our expert team does in-depth research to grade over 20 fundamental analysis components like scarcity, liquidity, team, and others.

To get the total fundamental grade, we have to assign weights to each component, and that’s where our machine learning models factor in.

We simulate 100 different weighting schemes, back-test each one, then we pick the best performing model in terms of accuracy, to then create new generations.

So with each generation, our models increase in accuracy.

We repeat this process over 500 times, resulting in picking the best performing fundamental analysis grade model from 50,000 different models.

We do the same process for our Technology grade, Technical Analysis grade, and the Overall grade, leading to a total of over 20 Million models PER DAY.

For example, consider the grading scheme of ETH for a day trader:

Our machine learning algorithms provide the optimal weighting scheme in calculating the overall grade.

And since no grade could reach 100%, we normalize the overall grade by taking a percentile grade.

For long-term value investors, we give more importance to our human expertise, so we calculate the overall grade by equally weighting the fundaments, technology, and technical analysis grades.

In addition to that, we also provide 30-day price predictions through our Multi-layer Perceptron Model (MLP), also known as a feed-forward artificial neural network.

Although the models will never be perfect, they get more accurate with more data, and we are gathering a massive amount of data daily.

At Token Metrics, you can see the accuracy of our predictive models and use the 30-day pricing predictions to craft your trades.

So with Token Metrics, you get the benefits of quant trading and the constant improvement of machine learning, all in one platform.

Without having to be a developer!

Did this answer your question?