SURVEY OF TIME SERIES-BASED AUTOMATED TRADING STRATEGIES IN STOCK MARKETS USING MACHINE LEARNING AND DEEP LEARNING

Abstract

Predicting the stock market and automated trading are both very difficult practices because of the financial markets'
nature, which is highly volatile, dynamic, and non-linear. Time Series-Based Automated Trading Strategies in Stock Markets are
popular because they can help capture the intricate, changing patterns of the market and allow investments to be made based on
data. The current study thoroughly reviews all the methods of classical and modern forecasting of financial time series, mainly
from the perspective of their applications in stock trading. Traditional statistical models such as AR, MA, ARIMA, and
ARCH/GARCH have provided foundational methods for modelling dependencies in stock prices, solving volatility problems, and
determining when to apply the stationary process. However, as global financial markets have become increasingly complex, deep
learning (DL) architectures, such as RNNs, LSTMs, CNNs, and hybrid CNN-LSTM models, have risen to the forefront as adept
tools for managing sequential dependencies and deriving features for stock market prediction. In addition, machine learning (ML)
methods such as supervised Linear Regression, KNN, SVM, and Naïve Bayes, as well as unsupervised K-Means, DBSCAN,
Gaussian Mixture Models, and Mean Shift, have collectively provided robust solutions for market segmentation and pattern
recognition. The study portrays the coming of age, the hurdles posed by non-stationarity, and the future of sophisticated learning
frameworks for improving forecasting precision and automated trading in today's financial markets by combining these
approaches.