A Deep Neural Network Approach to Predict Stock Prices Using Unconventional Data: Some Evidences from India

Prediction of the stock price is one of the major concerns in today’s need. Due to the pandemic, financial stability of the country was cryptic. It made investors in a predicament whether to invest or where to invest. Predicting the stock market helps to determine futuristic stock value of financial exchange. In this paper, authors have used Convolution Neural Network (CNN) and XGBoost, a class of Deep Neural Networks in predicting the stock prices. CNN and XGBoost identify predominant features among various other features without human intervention. A solitary approach is used for predicting the stock market. Here, the quantitative data is converted into images and is modeled with Convolution Neural Networks 2D and XGBoost. Authors contributed modeling synthetic image representation of quantitative data. Image representation of numerical data is advantageous as it can be loaded with many features in a single four-quadrant image consisting of four time series in a single image. The paper presents prediction accuracies of various companies by testing on live data. Results show that CNN 2D and XGBoost models provide over 90% accuracy using unconventional data.