Improvement of Data Mining Models using Forward Selection and Backward Elimination with Cryptocurrency Datasets

Abstract

Cryptocurrency is a digital currency not managed by a state or central bank, and transactions are peer-to-peer. Cryptocurrency is still considered a speculative asset and its price volatility is relatively high, but it is also expected to become an efficient and secure transaction tool in the future. The purpose of this study is to compare and improve the performance of the Data Mining Algorithm model using the Feature Selection-Wrapper with the Binance Coin (BNB) cryptocurrency dataset. The Feature Selection-Wrapper approach used is Forward Selection and Backward Elimination. The algorithms used are Neural Networks, Deep Learning, Support Vector Machines, and Linear Regression. The methodology used is Knowledge Discovery in Databases. The results showed that from a comparison using K-Fold Cross Validation with a value of K=10, the Neural Network Algorithm has the best Root Mean Square Error value of 10,734 +/- 10,124 (micro average: 14,580 +/- 0,000). Then after improving performance using Forward Selection and Backward Elimination in the Neural Network Algorithm, the best performance improvement results are shown by using Backward Elimination with RMSE 5,302 +/- 2,647 (micro average: 5,805 +/- 0,000).