Implementation of four machine learning algorithms for forecasting stock’s low and high prices

Today, several tools and statistical techniques can be used to find profitable trading opportunities, particularly when it comes to predicting stock closing prices. Yet, a few studies have been done on daily low- and high-price predictions which are useful for estimating support and resistance prices and improving the timing of stock purchases and sales. In this paper, we suggest combining machine learning algorithms with seven statistical features to enhance the forecasts of the low and high prices for the upcoming 5 days. In the experiment, the performances of linear regression, k-nearest neighbor, support vector machine (SVM), and the enhanced bidirectional LSTM (Bi-LSTM) were compared. Our findings showed that the Bi-LSTM surpassed the competition with a low RMSE of 0.018, a decrease in the error of over 58% compared to SVM, and a reduction of 51% compared to the unimproved LSTM. The four implemented algorithms are accessible online through a web application that displays the trend for the upcoming 5 days together with projected lines for the best-selling price (the high price) and the best-purchasing price (the low price).