2026-05-10T07:57:04.599Z http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai

ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai:3:9452 2024-12-03T20:10:47Z 3

Optimizing Long Short-Term Memory by Improved Teacher Learning-Based Optimization for Ethereum Price Forecasting, Chapter in LNDECT Lecture Notes on Data Engineering and Communications Technologies: ICMCSI 2023: Mobile Computing and Sustainable Informatics, Springer, volume 166 2023 http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai/record/3/9452 https://link.springer.com/chapter/10.1007/978-981-99-0835-6_9 M. Milicevic L. Jovanovic N. Bacanin M. Zivkovic D. Jovanovic M. Antonijevic N. Savanovic I. Strumberger Cryptocurrency has in the past decade established a certain foothold in modern economies. Wider adoption and an increase in popularity have led to many new currencies being created. Despite wider adoption cryptocurrencies’ value remains highly volatile, greatly affecting investment and trade decisions. The ability to forecast price fluctuations proves invaluable to both traders and speculators. However, the volatile nature of the cryptocurrency market makes casting accurate predictions challenging. This work proposed the use of a univariate times-series prediction approach based on long-short-term memory (LSTM) artificial neural networks for making accurate predictions based on real-world trading data. However, the performance of machine learning (ML) techniques such as the LSTM neural network is highly dependent on an initial set of control parameters that govern performance. To account for this, an improved version of the teacher learning-based optimization algorithm is proposed and tasked with selecting optimal parameters for LSTM network casting predictions. The proposed model has been validated on real-world data and put into a comparative analysis with other well-known metaheuristics applied to the same task. The overall outcomes where the proposed method obtained superior results with respect to the R2 value of 0.985, MAE of 0.014, and RMSE of 0.019 suggest that the novel proposed model has great potential when applied to cryptocurrency price forecasting. bookPart Springer, Singapore 125 139 10.1007/978-981-99-0835-6_9 LNDECT Lecture Notes on Data Engineering and Communications Technologies: ICMCSI 2023: Mobile Computing and Sustainable Informatics, volume 166