Kumar Aiswarya S. et al.; International Journal of Advance Research, Ideas and Innovations in Technology © 2019, www.IJARIIT.com All Rights Reserved Page | 320 ISSN: 2454-132X Impact factor: 4.295 (Volume 5, Issue 3) Available online at: www.ijariit.com A review on stock prediction using machine learning Aiswarya S. Kumar aiswaryaskumar97@gmail.com College of Engineering, Chengannur, Kerala Greeshma Merin Varghese greeshmamerinvarghese@gmail.com College of Engineering, Chengannur, Kerala Radhu Krishna R. mail2radhukrishna@gmail.com College of Engineering, Chengannur, Kerala Reshma K. Pillai reshmakpillai3@gmail.com College of Engineering, Chengannur, Kerala ABSTRACT The goal of this review is to describe the various methods used to predict the stock market, gold price and fuel price. The following paper describes the work that was done on investigating the application of regression, SVM, ELM, ANFIS techniques on the stock market price prediction. The report describes the various technologies with their accuracy level and efficiency in the test phase. It was found that support vector regression was the most effective out of the models used, although there are opportunities to expand this research further using additional techniques to incorporate the current affairs into the prediction features. Keywords— Stock prediction, Linear Regression, Fuzzification, SVM 1. INTRODUCTION The stock exchange is thought to be a fancy adaptive system that's tough to predict because of the big range of things that verify the day to day worth changes. We tend to try this in machine learning that tries to work out the link between variable quantity and one or a lot of freelance variables. Here, the independent variables square measure the options and also the variable quantity that we'd prefer to predict is that the worth. it's apparent that the options that we tend to square measure exploitation don't seem to be really freelance, we all know that the amount and outstanding shares don't seem to be freelance further because the price and also the come on investment not being freelance. This study aims to use completely different models to predict the worth changes and to judge the various model's success by withholding knowledge throughout coaching and evaluating the accuracy of those predictions exploitation legendary knowledge. These analysis considerations closing costs of the stocks. The model for the stock exchange was solely involved with the price for stocks at the top of a business day, high- frequency commerce is a locality of active analysis, however this study most popular a simplified model of the stock exchange. 2. MOTIVATION Stock market price prediction is an issue that has the capacity to be worth billions of dollars and is actively studied by the largest financial corporations in the world. It is a relevant problem because it has no clear solution. Several attempts can be made at approximation using many machine learning techniques. The project allows methods for real-world machine learning applications including acquiring and analyzing a large data set and using a variety of methods to train the system and predict potential outcomes. 3. METHODOLOGY 3.1 Linear Regression The regression method is finished through the sci-kit-learn machine learning library. This is often the core for the worth prediction practicality. There square measure some extra steps that have got to be done in order that the information will be fed into the regression algorithms and come plausible results. Especially, each coaching dataset should be normalized to a Gaussian usually distributed or normal-looking distribution between -1 and one before the input matrix is suited to the chosen regression model [1]. There square measure one or two necessary details to notice concerning the method the information should be pre-processed so as to match into regression models. Firstly, dates square measure usually portrayed as strings of the format ”YYYY-MM-DD” once it involves info storage. This format should be born-again to one whole number so as to be used as a column within the feature matrix. This is often done by victimisation the date’s ordinal worth. In Python, this is often quite easy. The columns within the information Frame square measure hold on as numpy datetime64 objects, that should be born-again to vanilla Python date-time objects that square measure successively born-again to associate whole number victimisation the to ordinal() constitutional perform for date time objects. Every column within the feature matrix is then scaled victimisation scikitlearn’s scale() perform from the pre-processing module. Mean absolute error methodology is employed to gauge the performance of the regression model.