Document Type



This item is available under a Creative Commons License for non-commercial use only


Computer Sciences

Publication Details

A dissertation submitted in partial fulfilment of the requirements of Technological University Dublin for the degree of M.Sc. in Computer Science (Data Analytics) 2021


The stock market prediction has been the subject of interest to various researchers and analysts due to its highly unpredictable nature and serves as a perfect example for time series forecasting. Over the years deep learning models such as Long-Term Short-Term Memory and statistical models such as Autoregressive Integrated Moving Average have shown promising results in predicting future stock prices. But the results from these models cannot be generalized as they fail to incorporate the dynamics of the market and influence of several external factors such as political, social, investor's emotion, etc on stock markets. Recently Facebook’s creation of the Prophet model solely for time series forecasting has been successful in fitting the trends and seasonality of the data accurately compared to vanilla models.

This research proposes a unique combination of the newly developed Facebook Prophet model and Attention-Based Long-Term Short-Term Memory model to predict the adjacent closing price of NIFTY 50 stocks to fit both the seasonality and non-linearity component of stock price data. Further to encompass both market and investor sentiments influencing stock prediction, data from five sources are collected from 01/01/2015 to 31/12/2019 namely historic stock price, technical indicators, news articles scraped from multiple news sources, and tweets collected from a verified Twitter account. To extract sentiments from unlabelled news and tweet data this research takes upon an unsupervised approach by implementing a pre-trained Bidirectional Encoder Representations from Transformers base uncased model. The proposed model is trained and validated on eight combinations of the dataset created by merging data from multiple sources and compared with the performance of the baseline Facebook Prophet model trained and tested with data from a single source i.e., historic stock prices. The proposed model resulted in the least Mean Absolute Percentage Error ranging from 3.3 to 7.7 for all the combinations of the data in comparison to the baseline model which achieved the highest Mean Absolute Percentage Error of 11.67.