Optimal Prediction of Bitcoin Prices Based on Deep Belief Network and Lion Algorithm with Adaptive Price Size

Abstract

This paper introduces a new bitcoin predictin model that includes three major phases: data collection, Feature Extraction and Prediction. The initial phase is data collection, where Bitcoin raw data are collected, from which the features are extracted in the Features Extraction phase. The feature extraction is a noteworthy mechanism for detecting the bitcoin prices on day-by-day and minute-by –minute. Such that the indexed data collected are computed regarding certain standard indicators like Average True Range (ATR), Exponential Moving Average (EMA), Relative Strength Index (RSI) and Rate of Change (ROC). These technical indicators based features are subjected to prediction phase. As the major contribution, the prediction process is made precisely by deploying an improved DBN model, whose weights and activation function are fine-tuned using a new modified Lion Algorithm referred as Lion Algorithm with Adaptive Price Size (LAAPS). Finally, the performance of proposed work is compared and proved its superiority over other conventional models.