Applying ridge regression and ANN to predict ICO price after six months

Abstract

ICO (Initial Coin Offering) is blockchain based crowdfunding venture to access capital. Investors buy this coin at an unissued time at an extremely cheap price. Then these coins are listed on the exchange, their price will go up extremely fast if the coin is good. Six months after ICO issuance is the period that investors expect to make a profit. The dataset of 109 ICOs is collected from reputable websites after data preprocessing. The correlation analysis between the 12 inputs shows that the data had multicollinearity problems, which led to skewed results of the multiple regression model. Overfitting occurs when using multiple regression models. To overcome the multiple regression model limitation, the Ridge regression method solves these ICO data problems of multiple regression models. The artificial neural network model solves the complex nonlinear relationships between inputs and ICO price. By tuning parameters to get the best performance according to three performance metrics Root Mean Square Error, Rsquares, and Mean Absolute Error, the ridge regression algorithm and artificial neural network achieved an accuracy of 63% to 92% and up to 98%, respectively in forecasting ICO prices with a test set of 3 ICOs that depend on the used metrics.

Keywords

• dự đoán ico
• hồi quy ridge
• hồi quy tuyến tính
• ico
• mạng neuron nhân tạo