PREDICTION OF STOCK PRICE TRENDS USING SENTIMENT ANALYSIS OF NEWS AND HISTORICAL STOCK PRICE DATA ANALYSIS WITH THE KHEDR, SALAMA, AND YASEEN (2017) DATA MINING MODEL

• Main Author: Manuel Hotasi, Abraham
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/79557
• Institution: Institut Teknologi Bandung

The stock price trend prediction is one of the most challenging topics to predict due to its dependence on numerous variables. Several studies have explored various methods for forecasting stock price trends, but much of this research has been conducted on foreign stocks. Based on a review of literature from the past five years in the field of data mining, the methodology proposed by Khedr, Salama, and Yaseen (2017), which combines sentiment analysis of news with historical stock price attribute data, achieved the highest accuracy (86.21%) in the training and testing phases compared to other studies. Therefore, the aim of this Final Project is to integrate the methodology used by Khedr, Salama, and Yaseen (2017) with stock data listed on the Indonesia Stock Exchange (BEI) to predict stock price trends that exhibit different characteristics (BBCA, BUMI, ASII). Additionally, this Final Project will also examine how the predictive model for stock price trends operates. The first step of this Final Project involves determining the polarity of financial news using a Naive Bayes classifier. This is followed by the second step, which involves combining the news polarity with historical stock price data to predict stock price trends using a KNN classifier. The first step of the Final Project, determining the polarity of financial news, successfully achieved prediction accuracies ranging from 90.8% to 94.8% in the training and testing phases and 90.6% to 97.3% in the validation phase. The second step, predicting future stock price trends, achieved prediction accuracies ranging from 71.9% to 90% in the training and testing phases and 42.1% to 64.6% in the validation phase.