While cryptocurrency and exhaust schemes grow, here's how to avoid scams with machine learning

[ad_2][ad_1]
W3Schools


In February 2018, with the increase in claims of virtual currency exchange frauds, the US Commodity Futures Trading Commission issued a notice to the public to create awareness. This was crucial because the incentives were so tempting that it is almost impossible to keep people away from this gold rush of the 21st century.

Risks involved in virtual currency

  • Most cash markets are not regulated or controlled by a government agency;
  • Platforms in the cash market may have no critical system guarantees, including customer protections;
  • Fluctuations in volatile market prices or abnormal flash arrests;

While governments around the world have been engaged in policymaking, Jiahua Xu and Benjamin Livshits, researchers at Imperial College London, have published a white paper discussing the anatomy of cryptocurrency Pump-and-dump schemes and how machine learning can be used to prevent such events in the future.



Advertisement


The researchers tracked the message history of over 300 Telegram channels between July and November of this year to identify pumping events. They analyzed the characteristics of the movement of coins on the market throughout the pump-and-dump process. And, they have developed an automatic learning model that works on a random forest algorithm to predict the likelihood of a possible pump event. The model confirms that market movements contain hidden information that can be used for monetary purposes.

Like Pump-And-Dump in 4 steps

  1. The organizer opens a channel accessible to a potential pumping group. Invite members by publicizing and publishing invitations on popular forums such as Reddit. Once the group exceeds 1,000 members, they are ready to pump.
  2. The organizer transmits the time and date of a future pump event. As time approaches, the administrator advises members on how to shop quickly and for how long to hold the coin to attract more users.
  3. The administrator announces the currency on a predetermined date and time. They use an OCR test model to circumvent and hinder the detection of the machine. During the first minute of the pump, the price of money generally increases, increasing in number.
  4. When the price of the coins approaches the maximum value, it begins to fall and the participants unload it or sell it to go out with full pockets. This trend continues until the price drops even below the original price.

In this case study, the researchers targeted BVB, an obscure currency listed on CoinMarketCap. This coin was launched in 2016 by supporters of the German football team, Borussia Dortmund.

The total volume of purchases measured in BVB is 1,619.81 thousand BVB, the amount of sale 1,223.36 thousand BVB. This discrepancy between the sales volume and the purchase volume indicates a greater commercial aggressiveness on the part of the purchase.

Predictive modeling: a ML approach

Features included in the model

The table above illustrates the key features that have been used to train the model. Due to the ease of standardizing data and due to its high pump-and-dump frequency, researchers focused on predicting coins pumped into cryptopenia.

On average, there are 358 coin changers for each pump, one of which is the real pumped coin. The number of coins considered varies for each event due to constant listing / cancellation activities on the scholarship list. The complete sample contains 47,787 pump-coin observations, of which 133 pumped cases, 15 representing 0.3% of the entire sample population. The sample is apparently heavily distorted to the unmanned class and must be handled with care during modeling.

To avoid over-processing, sample data are divided into three series of data in chronological order between July and October; generating 27,759 data points in the first set, of which 78 are pumped cases. The validation set consists of 10,106 data points, including 28 pumping cases. And the test set has 9,755 points with 27 pumped cases.

A random forest with stratified sampling for the classification was chosen and a generalized linear model (GLM) for regression of the logit.

Due to the strongly unbalanced nature of the sample, when using the RF, the model always includes TRUE cases when starting the sample to create a decision tree.

The RF1 model remains true to the original TRUE / FALSE ratio of the sample, with 0.3% of TRUE contained in each tree sample. RF2 and RF3 increase the TRUE / FALSE ratio of 1.2% and 6%, respectively.

Considering that, LASSO (less absolute withdrawal and selection operator) the regularization is applied to GLM models to avoid problems deriving from distorted distribution.

Both the random forest model and GML are able to predict if a given coin will be pumped as a probability between 0 and 1. In terms of the F1 measurement, the RF models, in general, appear superior to the GLM models with both the sample of training that the validation sample.

At each pumping event, the researchers controlled the normalized vote of the currency that violated the predetermined threshold limit. They then bought the coin an hour before the announcement based on the model's prediction. With all the coins purchased, the investment, measured in BTC, on each coin is proportional to its vote provided by the random model of the forest.

The results show that the model suggests the purchase of 6 coins, of which 5 are actually pumped.

This study

  • It is the first of its kind of pump-and-dump schemes in the world
  • Show that pump-and-dump activities are much more common than they previously believed. Specifically, approximately 100 Telegram pump-and-dump channels coordinated on average 2 day pumps generate an aggregate artificial trade volume of $ 7 million per month.
  • Aided in the development of a predictor that, given a pre-pump announcement, can predict the probability that each coin will be pumped with an AUC (Area Under Curve) of more than 0.9 in both sample and out-of-sample.
  • It is a simple trading strategy that, on the basis of historical data, returns a yield of 80% over a period of three weeks, even in the presence of precise assumptions.

when Satoshi Nakamoto has proposed its peer-to-peer electronic payment system, has envisioned a reliable and foolproof transaction service that allows the two parties to trade without the involvement of any financial institution. Like any other profitable business, this one has attracted stupid and scams where it thrives at the expense of others. Although the virtual currency itself is secure with cryptography, new methods have been devised to sensitize the system as discussed above. And automatic learning could be such a solution that should be used to cement cracks.

Provide your comments below

Comments

[ad_2]Source link