Stock Price Prediction: Fuzzy Clustering-Based Approach

Introduction

Forecasting the stock market index and index movements is one of the most challenging time series analysis obstacles. Investors use two types of analysis, fundamental and technical analysis, before investing in stocks. With fundamental analysis, investors decide whether to invest, taking into account indicators such as the stock's actual value, the political climate, the industry's performance, and the economy. In technical analysis, the evaluation of stocks is provided by using statistics created by market movements such as historical values and transaction volumes (Pabuççu, 2019).

Investors are beginning to rely on forecasting systems to make critical business decisions. There is a lot of research done in this field, but no complete solution has yet been found. Difficulty in predicting the stock market depends not only on the influence of social, political and economic reasons but also on a vast amount of historical data about the stocks and currency (Iqbal et al., 2013). This study consists of many states of the art machine learning techniques to find optimum solutions for predicting stock values.

In the literature, there are two types of approaches to forecasting a stock price. These are qualitative and quantitative approaches. The quantitative approach uses past stock prices, such as the closing and opening price, the amount exchanged, neighbouring closing rates of the stock etc., to forecast the stock's future price. In the qualitative approach, the analysis is based on external factors: economic and political factors, company's identity, company's and general market situation etc. In this approach, textual information published in the magazine or web and social media blogs are used, written by economic experts (Hur et al., 2006).

This study aims to predict bank stocks: AKBNK, GARAN, HALKB, YKBNK and VAKBN in BIST 30 index. For that reason, the authors use the last three years' BIST 30 index values for predicting those stocks values. Also, the fuzzy clustering technique is applied to the dataset to find stocks in BIST 30 index and have similar characteristics with predicted bank stocks. The main reason for this approach is to test whether enlarging the data set to be used in the prediction phase positively affects the performance of the algorithms.

The authors applied classical machine learning algorithms such as Random Forest, Support Vector Machine etc. But, ensembling types of machine learning algorithms such as XGBoost, Catboost, LightGBM and GBDT produced more promising results with different hyperparameters for the prediction. Besides, the most successful models with their hyperparameters were ensembled according to their error rate's reciprocal for better prediction performance. These results were compared with the traditional methods, and it is discussed in the methodology section.

In this paper, the details of the methodology are provided in Section 2. Section 3 discusses the result of the findings, and Section 4 concludes the article.