Machine learning applications for time series analysis

Abstract

In this studying, involves doing a range of tests using time series data sets collected from stock markets (BIST30, BIST100, Apple) and cryptocurrency marketplaces. Statistical analysis and artificial intelligence models are employed to investigate various data sets inside the studies, and the findings are subsequently analysed. The main objective of the study is to provide a valuable contribution to academic research and offer practical advantages to market investors. Consequently, the researcher has thoroughly examined the current models and studies in the literature and has chosen the most suitable artificial intelligence models (ARIMA, SARIMA, GARCH) for the thesis study. The paper extensively discusses and applies these concepts within its scope. The study's findings indicate that no existing framework can accurately forecast the time series-dependent pricing of crypto assets traded on stock exchanges and crypto exchanges. These conclusions are based on the results gained from the experiments conducted. The primary factors contributing to this unpredictability can be ascribed to market price volatility and the fact that price variations generate outcomes regardless of time. Subsequent investigations might prioritise the utilisation of additional data sources to enhance the existing time series data, hence enhancing the precision of prediction outcomes. Incorporating supplementary information such as macroeconomic indicators, sector-specific data, geopolitical events, and social media sentiment can augment the precision of prediction models. This thesis study offers essential insights into the predictability of financial time series. The present pricing and daily price changes alone are inadequate in providing credible predictions. This is because elements such as seasonality, seasonal variability, and periodic trends, which are stochastic in nature, make the prediction process more complex. The thesis clearly demonstrates the constraints and difficulties encountered in financial market analysis as described in the literature, with the assistance of data derived from both literature-based research and experiments. The statistical methods used and the data gained in this study serve as an initial investigation, with the goal of establishing a methodological basis for future studies and opening up possibilities for further research in many areas. This study is expected to serve as a benchmark for future market researchers and academics conducting research in this subject.