Cryptocurrency analysis and predictions using AI and big data

elavmunretea said:

Love the site. Not sure about the domain though

It would be nice if the charts could have horizontal overlays to see the price a bit easier, so I don't have to look at the left side all the time.

It's also slightly weird that you can move the price without moving the other data.

Click to expand...

healzer said:

I hope this makes some sense

Click to expand...

elavmunretea said:

Yes, I understand that, but is there a way to move both of them proportionally?

Click to expand...

healzer said:

Hey all

Yesterday I had an idea on how to realize @Ensili 's idea of categorizing crypto coins into "stages".
Every coin has a few features that make it unique in the marketplace:

1. The hype around it (be it news or social media).
2. The traded volume.
3. And its price, which is, related to and influenced by, the two above.

So my thought was to visualize all three features for each coin, at a certain point in time (e.g. the past 24 hours).
We can only (practically) work up to three dimensions, but I have two types of hype: news and social media.
A basic workaround for this problem would be to multiply news mentions by social mentions (unless any is zero).
But if you think about it, we would be very interested in seeing news and social separated.

So I had to introduce a ""4th"" dimension.
This was realized by changing the size of a data-point according to the traded volume (larger dot means more trades).

Here's my result:

You can play around with this plot yourself on this link: http://cryptanal.com/stagesPriceHypeVol3d.php
A few notes about the data:

• In this example we have used data from the past 24 hours.
• The price of a coin is the average of prices in the past 24 hours.
• The news hype is the sum of news mentions in the past 24hrs.
• The social hype is idem ditto ^.
• Volume: some coins (e.g. XRP) have a traded volume24h of hundreds of millions, while others (e.g. BTC) are between 80k and 180k. So to display the size of a dot properly we have a formula like: dot_size = log( vol24h**2 )/log(6) . If we don't take the log, we'll end up with crazy large dots that take the entire plot size.
  
  
• x axis is the price
  y axis is news mentions
  z axis is social mentions

You will see that Bitcoin is at the highest stage, followed by ETH and then LTC.
We also see that BTC, ETH and LTC are lining up, and it seems that IOT and/or XMR will be next to line-up.

It's also very cool to see which coins are heavily traded (e.g. DOGE) but are very low on social/news mentions.
XRP is the most heavily traded one, and floats in its own "zone", away from the masses:

A next step for "staging" coins would be to show their respective paths.
I mean, it would be even more meaningful if we could see how a certain coin (e.g. LTC) evolved in the past 3 months.
Once we can visualize its path, we could see which coins have followed a similar path as LTC did and might go to where LTC is now (within X weeks/months).

Have a great day all!
- Ilya

Click to expand...

• Sequence length: 20 intervals
• Features: 5 (price, volume24, social hype, sentiments from news articles, sentiments from social media)
• Neurons: 2
• Epoch: 500
• batch size: 20

elavmunretea said:

You should also look at time of day & day of the week.

Some coins are more active on certain time zones and go down on specific days like weekends.

Click to expand...

mindmaster said:

Any way of making prediction for ICOs, as well ?

Click to expand...

healzer said:

ICOs are not yet supported and will require different analyses.
I may look into that very soon.

Click to expand...

elavmunretea said:

I am also looking at automating ICOs. Really hard (pretty much impossible) to predict sucess automatically.

I know that the Cindicator bot offers ICO rating, and you could maybe use ICOBench's API but it's honestly just easier to spend an hour or so researching each ICO yourself.

Click to expand...

healzer said:

I'll definitely try to use them.

Click to expand...

Dave Dubauskas said:

WOW! Well done!
I have not read all the posts so please disregard this suggestion if it has already been posted. If I could suggest it would be useful to included social indicators for trading. What do I mean? Well, for instance, there is a bot that tracks "mentions" of crypto coins on twitter. Google trends also indicate interest in buy or selling coins. I read (somewhere/sometime) ago an article that listed and discussed the "social indicators" of trading... but of course, I can't find it now to post here?! You may wish to keep it separate at first and study the correlations to the actuals? You will decide of course whether there is merit in including some the indicators in your price predictions. Again well done and thank you for work. Cheers.

Click to expand...

Bel1616 said:

Hey as always good job! I'm a big fan of this project, will you do cryptocurrency only or add on the big stocks in the future?

Click to expand...

healzer said:

Hi guys

I have got something exciting for you.
Today and yesterday I have invested all my time implementing the auto-prediction system.
This new feature automatically generates predictions for every coin.
Right now it makes 4 predictions per coin, for 4 different intervals: 30min, 60min, 12hrs and 24hrs.
Basically It predicts 20 intervals into the future depending on the interval you choose.

Here are a few screenshots before I go into more details.

IOTA, 30min intervals:

BTC: 30min, 12hr and 24hr intervals:

A few technical notes:
Every 30 minutes the system generates a prediction per crypto and per interval.
There is a reason for generating a prediction for each possible interval: remember that we noticed that the predictions are only accurate for about 5 - 12 intervals into the future. So for instance if we would want to predict 20 days ahead, we don't want to use 30min intervals to generate 960 prediction-points. So instead we train the model on the 24hr-based data and make 20 prediction-points from it.

Feel free to use the tool here: http://cryptanal.com/predictionsChart.php
When you are going to use the prediction tool, you may see an empty chart -- this is because it has not generated a prediction (yet) for the datetime you selected.
Simply click "prev" and/or wait until the new prediction is made.

You can use the "prev"/"next" button to go to the previous/next interval. I added the prev/next buttons to see which predictions were made in the past, so we can compare history against reality. However, only a limited number of predictions have been recorded until now (only the most recent ones), so you will not find a prediction made e.g. 10 days ago.

On the gif below you can see how the prediction evolves when pressing "next" (I started at some historic datetime):

Notice that each prediction looks (slightly) different from the previous one.
This is because the neural network is re-adjusted with the new data (continuous learning) every interval.

In one of my earlier posts I explained that there are a bunch of parameters/configurations for the prediction algorithm.
I have not yet figured out which parameters are the "optimal" ones, this is not a trivial task but it will be done in the near future.
So the current algorithm uses the parameters which gave me the best results thus far:

• Sequence length: 20 intervals
• Features: 5 (price, volume24, social hype, sentiments from news articles, sentiments from social media)
• Neurons: 2
• Epoch: 500
• batch size: 20

Regarding the price and volume: I may not have mentioned this before, but the price we record is the average price of a cryptocurrency across all mainstream exchanges. The same is true for the traded volume, the system takes the average of the traded volume across all exchanges. Maybe in the future I will implement per-exchange charts which is useful to very serious traders.

Apart from implementing the prediction tool, I have patched a few bugs.
The most important one is displaying more digits after the decimal point/comma. Previously it was rounded to 2 digits, but I noticed that some coins are worth less than one US Dollar (e.g. NXT):

Thanks for reading, I love all of your support and feedback!!!
Would love to hear your thoughts on this one as well.
- Ilya

Click to expand...

ses said:

Based on the last gif where you show how your predictions evolve and the real stats proceed over time it seems like they are not even close to be accurate.

I am wondering if flipping a coin for each interval with the options for 'up' and 'down' would result in more accurate predictions.

Click to expand...

healzer said:

Great question!
On first sight it may appear to be inaccurate, but there is a reason for that.
Allow me to explain how it works.

Here is the gif @ses is referring to:

A certain frame on this gif is the chart below.
It was generated for the interval 22:00 --> 22:30 (GMT+1).
And it shows that the price is most likely to crash/go down from 11.5k to 10.9k:

But after a few hours its outcome has been adjusted.
It learned that the price didn't crash so the outcome started to change:

But if we look on the screenshot below, and compare it with the prediction above from 06:00 to 12:00 , it appears to be a pretty good prediction of how the price evolved. At first it went up a little in the first few hours and then slowly started going down. But right now (as I am writing this post), it indicates the price may crash/go down again:

*edit* Here's an easier way to look at what I mean:

It's important to remember that the screenshots above all use 30min intervals. So the neural network is only trained on plots at 30min intervals, it's not very good at predicting e.g. +8 hours into the future because that's already 16 intervals it has to predict, while we noticed it's only good for up to 5-10 intervals into the future.
So to tackle this problem you may want to use hourly/12h/daily intervals instead:


Below are two screenshots which use hourly intervals.
The first one makes a prediction on Jan 27, 22:00 for the next 20 hours:

The second one I have pulled from the site right now and shows how the price evolved, including its prediction:

Now we can compare the first prediction with the 2nd screenshot's actual data.
Now find the x-tick "Jan 27, 22:00" on both screenshots and look how the prediction differs from actual data.
We do see that at 02:00 --> 06:00 the prediction showed it would go down, the reality was less worse than indicated, we did have a dip at 01:00 and at 04:00.
Then from 06:00 until 14:00 the price went up and both areas look very similar, so the prediction is pretty accurate for this zone.
Here is a better illustration of what I mean:

And finally, for the 12hr interval we have the following (see image below).
It predicted the price to go up to 11.5k and so it did in reality.

On first sight it appears that hourly/12h intervals are much more accurate than 30min intervals.
We don't have enough data yet to analyze the 24hr intervals, but in due time we'll dig deeper.

I hope this clears things up a bit

Click to expand...