Page 66 - 2022

industry? It is also another question that creates lots of opportunities in the area of
research to create models with accuracy that help answer these questions.
Nowadays, stock movement prediction has been the main source of interest for
numerous researchers due to its random nature, complexity and economic rewards. The
Current study is based on an idea generated by the researcher who has been trading in
the share market by just taking tips from friends, without any further analysis for the last
five years. Sometimes, the researcher has born big loss of his money. Hence, the
researcher has decided to make a model which can help investors to make any type of
trading decisions in the stock market and protect the investors from future wealth
destruction. The main objective of the proposed study is to find short-term trading
opportunities by using the sentimental and technical parameters which can further be
used by retail investors for taking trades in the stock market for making wealth.

In the present study, three important industries namely pharmaceutical, real estate and
bank are taken as samples for three stocks in the benchmark. This study covers data
from more than 5 Lakh feeds taken from social media from 1st January 2018 to 31st July
2021 to train the model. The feed will be used to extract the features of importance in
prediction through the text mining process. Furthermore, feature extraction is done
through a sensitivity score algorithm to calculate the sentiment score, impact and
semantic classification of the users’ views.
The total data set is split into two parts as training and testing sets. The K-fold
methodology is used for the same. Six machine learning methods, namely KNN, Decision
Tree, Neural Network, SVM, Logistic regression, and Random Forest, are used to create a
hybrid model of prediction buy and sell call for the particular equity share. The main
contribution of the study is to increase accuracy from 73 to 89 percent to more than
89.41 percent in each equity share prediction of buy or sell call. This hybrid model
specifically focuses on buying part to protect investors from future wealth destruction
through sentiment analysis.
The proposed hybrid model is based on customized ensemble stacking technique in
which sentimental and technical parameters are used as input parameters for the
training and testing of the model. It consists of two layers, in first layer, six heterogeneous
machine learning algorithms such as KNN, decision tree, support vector machine, logistic
regression, neural network and random forest are used as base learners for the training
process by using the training data. In second layer, the output of base learners is
aggregated by using the meta-learner to make the final prediction.
The entire thesis is divided into five chapters to discuss what, how, and why for a detailed,
in-depth discussion. In chapters 1 and 2, introductions to the study are covered with the

35