Financial data forecasting has historically been a significant domain with a wide array of consequences and applications. In today’s tech driven era, it has facilitated the prevalence of data-driven decision making, reduced uncertainties in the financial arena and propelled monetary profits unprecedentedly. Numerous techno-social algorithms and technologies have been employed to facilitate financial data forecasting. Modelling these forecasters is not easy since financial markets are unpredictable, noisy, non-linear and volatile. However, the implementation of machine learning can prove to be vital. Many researchers across the globe have attempted to use sophisticated machine learning models to predict stock prices and stock price movements accurately.
In order to adopt a more holistic approach to stock price prediction, this talk will cover prediction of of selected small-cap and large-cap stocks/ indexes. Furthermore, in this talk, we will explore and experiment different combinations of input features which generate lowest error. Finally, various machine learning and deep learning models are explored and evaluated to discover the most accurate and reliable model.
-Introduce myself and my background
-Explain my motivation behind choosing this topic
-Introduce the topic and its significance
OVERVIEW OF METHODOLOGY
-Data preprocessing and analysis
-Creation of new features: Technical Indicators and Sentiment Analysis
-Feature Engineering: Principal Component Analysis (PCA) and Recursive Feature -Elimination (RFE)
-Machine learning/ deep learning: LSTM, LR, SVM etc.
-Introduce the kinds of data collected and the reason behind doing so
-Numeric Data: Historic prices from Yahoo Finance
-Non-Numeric Data: News Headlines from Reuters.com
DATA PREPROCESSING AND ANALYSIS
TECHNICAL INDICATORS AND SENTIMENT ANALYSIS
-Technical indicators: Simple Moving Average (SMA), Exponential Moving Average (EMA), -Relative Strength Index (RSI)
-Polarity scores using VADER’s Sentiment Intensity Analyzer
-Subjectivity scores using TextBlob
-Headline embedding using Google’s Universal Sentence Encoder
-Dimensionality reduction using Principal Component Analysis (PCA)
-Feature Selection using Recursive Feature Elimination (RFE)
LONG SHORT TERM MEMORY
-What is LSTM?
-Why do we pick LSTM over other algorithms?
-Optimization of hyperparameters and parameters
RESULTS AND DISCUSSION
-Discuss the results
Speaker: Ms. Anshika Rajiv / Hong Kong / Computer Science Student at the University of Hong Kong (HKU) - Website, GitHub, Facebook Language: English Date and Time : October 9, 2021 / 11:15-11:45 (UTC+8)