Hakan Gunduz

Borsa Istanbul (BIST) daily prediction using financial news and balanced feature selection

The direction of Borsa Istanbul 100 Index (BIST100) open prices is predicted.A feature selection method, called Balanced Mutual Information (BMI) is proposed.BMI is able to deal with the class imbalance problem through oversampling.BMI is compared with Mutual Information and Chi-square based feature selection.BMI achieves higher macro-averaged F-measure than the other methods using less features. In this paper, a novel method is proposed to predict the direction of Borsa Istanbul (BIST) 100 Index (BIST100) open prices using the news articles released, as well as the price data, from the day before. Although English news articles have been used for market-prediction before, to the best of our knowledge, Turkish news articles together with prices have not yet been used to predict the Turkish markets. Turkish text mining techniques are applied on news articles to form feature vectors for each trading day. The feature vectors are assigned three labels based on the direction of the price change from the closing price of the day before and whether the change is significant. News articles are represented using high dimensional features, some of which could be noisy or irrelevant for prediction. There is also the scarcity of training data. Therefore, this study incorporates feature selection methods to select features that could improve classification performance. By its nature, significant positive or negative changes in stock price happen much less than non-significant changes, resulting in an imbalanced data set. Most feature selection methods in literature aim to reduce the classification accuracy. However, for imbalanced datasets, other measures, such as macro-averaged F-measure need to be considered. The paper proposes a feature selection methods that is able to deal with the class imbalance problem through oversampling of the minority classes and consideration of an ensemble of selected features. In order to decide on importance of features, as the relevance criterion for each feature, the proposed methodology uses mutual information which can detect nonlinear dependencies between variables. Therefore, the proposed feature selection method is called Balanced Mutual Information (BMI) feature selection method. Experiments were performed based on news articles provided by two different news sources: Public Disclosure Platform of BIST and financial news websites. It was shown that, using Balanced Mutual Information feature selection method, the significant changes in the BIST100 Index were predicted with an accuracy of 0.74 and a macro-averaged F-measure of 0.68. The BMI feature selection method was compared with Mutual Information and Chi-square based feature selection methods and it was found out that BMI method results in higher performance using a smaller number of features.

Stock market direction prediction using deep neural networks

In this study, the daily movement directions of three frequently traded stocks (GARAN, THYAO and ISCTR) in Borsa Istanbul were predicted using deep neural networks. Technical indicators obtained from individual stock prices and dollar-gold prices were used as features in the prediction. Class labels indicating the movement direction were found using daily close prices of the stocks and they were aligned with the feature vectors. In order to perform the prediction process, the type of deep neural network, Convolutional Neural Network, was trained and the performance of the classification was evaluated by the accuracy and F-measure metrics. In the experiments performed, using both price and dollar-gold features, the movement directions in GARAN, THYAO and ISCTR stocks were predicted with the accuracy rates of 0.61, 0.578 and 0.574 respectively. Compared to using the price based features only, the use of dollar-gold features improved the classification performance.

Intraday prediction of Borsa Istanbul using convolutional neural networks and feature correlations

We have extracted different types of indicator, price and temporal features.Previous instances and correlation between features are used to design CNN.We predict the hourly direction of 100 Stocks Borsa Istanbul Stock Market.Proposed method outperforms the CNN that uses randomly ordered features.On average we perform 56.3% Macro Average F-Measure rate on 100 stocks. Stock market price data have non-linear, noisy and non-stationary structure, and therefore prediction of the price or its direction are both challenging tasks. In this paper, we propose a Convolutional Neural Network (CNN) architecture with a specifically ordered feature set to predict the intraday direction of Borsa Istanbul 100 stocks. Feature set is extracted using different indicators, price and temporal information. Correlations between instances and features are utilized to order the features before they are presented as inputs to the CNN. The proposed classifier is compared with a CNN trained with randomly ordered features and Logistic Regression. Experimental results show that the proposed classifier outperforms both Logistic Regression and CNN that utilizes randomly ordered features. Feature selection methods are also utilized to reduce training time and model complexity.

An Extension to Local Network Alignment using Hidden Markov Models (HMMs)

Local alignment is done on biological networks to find common conserved substructures belonging to different organisms. Many algorithms such as PathBLAST (Kelley et al., 2003), Network-BLAST (Scott et al., 2006) are used to align networks locally and they are generally good at finding small sized common substructures. However, these algorithms have same failures about finding larger substructures because of complexity issues. To overcome these issues, Hidden Markov Models (HMMs) is used. The study done by (Qian and Yoon, 2009), uses HMMs to find optimal conserved paths in two biological networks where aligned paths have constant path length. In this paper, we aim to make an extension to the local network alignment procedure done in (Qian and Yoon, 2009) to find common substructures in varying length sizes between the biological networks. We again used same algorithm to find k-length exact matches from networks and we used them to find common substructures in two forms as sub-graphs and extended paths. These structures do not need to have the same number of nodes and should satisfy the predefined similarity threshold (s0). The other parameter is the length of exact paths (k) formed from biological networks and choosing a lower k value is faster but bigger values might be needed in order to balance the number of matching paths below s0.

Glasses detection in face images using histogram of Oriented Gradients

Glasses detection is one of attractive tasks in image processing since it increases the performance of face recognition systems. In this study, we aimed to detect the glasses on face images automatically. In order to do this, we trained a classifier with Labelled Faces in the Wild Home (LFW) dataset to decide whether a person wear glasses or not on face images. Before classification process, image patches are extracted from aligned face images and a preprocessing was performed on them. After preprocessing step, feature vectors are formed with Histogram of Oriented Gradients (HOG) method from image patches. Due to high dimensionality of the feature vectors, dimensionality reduction was done using Principal Component Analysis (PCA). The dimension-reduced feature vectors were splitted into training set and test set. With training set images, Support Vector Machines (SVM) classifier was trained and the model parameters were defined. The classifier performance was evaluated with test set images and nearly 93% accuracy rate was achieved.