Stock Price Prediction using Hybrid Approach

In this paper we have focused our attention to create a hybrid approach providing optimal results than other currently working models and approaches. Our approach had r2score of 0.97 also the error values like MAE and MSE were 0.018. It is a well-known fact that stock market is volatile and dynamic in nature being influenced by many factors which are uncountable. In this era of technological advancements many machine learning models working like human brain have been developed like Artificial neural network. Here we collected data from banks listed on National Stock Exchange(NSE) spanning more than 13 years. The proposed solution in this paper is a comprehensive approach as it contains pre-processing, making use of feature engineering methods, building a customized approach of artificial neural network and principal component analysis. Stock prices are heavily depended on numerous features, among them only few ere available in our collected dataset. We generated new real-life features i.e feature expansion creating a pool of features influencing stock prices. By using PCA on the newly formed data the features were reduced resulting into dimensionality reduction. Using PCA more correlated features affecting the stock prices are identified which would eliminate the features that does not influence stock price to certain extent. These newly formed reduced features acts as an input for the artificial neural network making the model faster and making use of all the necessary features. The results of our method is finding an optimal number of Principle components and the necessary layers of Artificial Neural Network (ANN) for better price prediction. In here we did a comprehensive analysis on most used machine learning models for stock price prediction based on the error rates and our proposed hybrid approach outperforms them due to comprehensive feature engineering.