Core Technology - AI

Model building

The time series is an ordered collection of measurements at regular intervals, such as daily stock prices or weekly sales data. The time-series modeling approach assumes that history will always repeat itself. Even if not exactly the same but close, so it is enough to make better decisions about the future by studying the past. At present, Banksea has completed the construction of the NFT valuation model on Solana, Moonbeam and part of Ethereum. It is constantly keeping optimizing and updating. We will provide the NFT valuation for more chains soon.

model-building

Model Evaluation

There are many methods to evaluate the time series models. Commonly used time series model evaluation methods are as follows:

model-evaluation

Since the prices of different NFT collection vary greatly, it is not appropriate to compare them directly by deviation. Therefore, we use MAPE to comprehensively evaluate the model and refer to other evaluation indicators.

Mean Absolute Percentage Error (MAPE), also known as Mean Absolute Percentage Deviation (MAPD), is a measure of the predictive accuracy of a forecasting method in statistics. It usually expresses precision as a ratio defined by the formula:

$$MAPE={\frac {100\%}{n}}\sum _{i=1}^{n}{\frac {y_{i}-\hat{y}_{i}}{{y}_{i}}}$$

$y_i$ is the actual value and $\hat{y}_{i}$ is the predicted value. Their difference divided by the actual value $y_i$. The absolute value of this ratio is the sum of each predicted time point divided by the number of fitting points, $n$.

More model evaluation methods are as follows:

• $ R^2$ - coefficient of determination

$$R^{2}=\frac{S S R}{S S T}=\frac{\sum\left(\hat{y}_{i}-\bar{y}\right)^{2}}{\sum\left(y_{i}-\bar{y}\right)^{2}}$$

• MAE - Mean Absolute Error

$$M A E=\frac{1}{n} \sum_{i=1}^{n}\left|y_{i}-\hat{y}_{i}\right|$$

• MSE - Mean Squared Error

$$M S E=\frac{1}{n} \sum_{i=1}^{n}\left(y_{i}-\hat{y}_{i}\right)^{2}$$

• MedAE - Median absolute error regression loss

$$\operatorname{MedAE}=\operatorname{median}\left(\left|\hat{y}_{1}-y_{1}\right|, \ldots,\left|\hat{y}_{n}-y_{n}\right|\right)$$