46 Ultima Infosys : Jurnal Ilmu Sistem Informasi, Vol. 13, No. 1 | June 2022 ISSN 2085-4579 Finding Features of Multiple Linear Regression On Currency Exchange Pairs Raymond Sunardi Oetama 1 , Ford Lumban Gaol 2 , Benfano Soewito 3 , Harco Leslie Hendric Spits Warnars 4 1,2,3,4 Doctor of Computer Science, Bina Nusantara University, Jakarta, Indonesia raymond.oetama@binus.ac.id fgaol@binus.edu bsoewito@binus.edu spits.hendric@binus.ac.id Accepted 30 June 2022 Approved 13 July 2022 Abstract— Due to the prospects for financial gain, forex is always attractive to many people. However, because forex market analysis is not simple, a computer is needed to assist in creating predictions using features that are understandable to people. This study employs the Multilinear Regression technique to identify these kinds of features. The features and prediction target have a very strong correlation with the lowest RMSE is 0.00408 and the highest R2 is 0.99477, the prediction quality is quite outstanding. The outcome will help academics in the forex field use machine learning algorithms to make better predictions. Index Terms—Features; Forex; Machine Learning; Multilinear Regression; Predictions; I. INTRODUCTION People trade for money by exchanging products for goods, goods for money, and, more recently, money for money. Forex trading is a type of trading in which one country's currency is swapped for another country's currency [1]. Not all currencies, however, are exchanged. EUR (European Union currency), GBP (British pound sterling), AUD (Australian Dollar), NZD (New Zealand Dollar), USD (United States Dollar), CAD (Canadian Dollar), CHF (Swiss franc), and JPY (Japanese yen) are the most widely traded currencies. Currency pairs are used in forex trading; for example, GBP and USD are coded as GBPUSD, EUR and USD are coded as EURUSD, and so on. A Currency trader must have the knowledge and analytical capabilities of the transactions that take place in the forex market to generate a profit. If a trader takes a BUY action when the price rises, the trader will profit. Meanwhile, traders who take SELL actions will benefit as prices fall [2]. The currency market, on the other hand, is difficult to analyze [3]. This is since manual analysis is limited, but forex prices move in milliseconds. As a result, computers must assist humans in analyzing forex price fluctuations. Because the parameters utilized in the analysis may be no longer relevant in today's world. The algorithmic trading approach is one way to undertake forex price analysis with the use of an algorithm [4]. Multiple Linear Regression is one of the algorithms employed in this strategy. Multiple Linear Regression is, in fact, an old algorithm. Even so, its capacity to explain which features influence prediction results, on the other hand, is undeniable [5]. This advantage will be used in this study to identify traits that are extremely useful in decreasing forex price forecast errors. The opening price, the highest price, the lowest price, and the closing price are the four original prices that are used to express forex price fluctuations in Japanese candlesticks [6]. As a result, this research contributes to the discovery of traits that contribute significantly to the reduction of forex price forecast errors. The closing price is used for prediction in this study since it occurs at the end of each day’s transaction. The purpose of this study is to find features that are highly connected with prediction targets that are relevant to the current situation of the forex market and to reduce mistakes in forex price predictions. Whilst limitation of this study is other pairs that aren't limited to GBPUSD and EURUSD require more investigation. Only applies to feature groups, not to feature merging.