Luis E. Zárate

Applying Artificial Neural Networks to prediction of stock price and improvement of the directional prediction index – Case study of PETR4, Petrobras, Brazil

Abstract Predicting the direction of stock price changes is an important factor, as it contributes to the development of effective strategies for stock exchange transactions and attracts much interest in incorporating variables historical series into the mathematical models or computer algorithms in order to produce estimations of expected price fluctuations. The purpose of this study is to build a neural model for the financial market, allowing predictions of stocks closing prices future behavior negotiated in BM&FBOVESPA in the short term, using the economic and financial theory, combining technical analysis, fundamental analysis and analysis of time series, to predict price behavior, addressing the percentage of correct predictions of price series direction (POCID or Prediction of Change in Direction). The aim of this work is to understand the information available in the financial market and identify the variables that drive stock prices. The methodology presented may be adapted to other companies and their stock. Petrobras stock PETR4, traded in BM&FBOVESPA, was used as a case study. As part of this effort, configurations with different window sizes were designed, and the best performance was achieved with a window size of 3, which the POCID index of correct direction predictions was 93.62% for the test set and 87.50% for a validation set.

The use of artificial neural networks in the analysis and prediction of stock prices

In recent years there has been a significant growth of interest in the incorporation of historical series of variables related to stock prediction into mathematical models or computational algorithms in order to generate predictions or indications about expected price movements. The objective of this study was to utilize artificial neural networks to predict the closing price of the stock PETR4 which is traded on BM&FBOVESPA. Three stages were used to generate the prediction: obtainment of the samples, pre-processing, and prediction. 32 different configurations were created by varying the window size and prediction horizon. The best performance was obtained with 5 days of quotes and a prediction horizon of 1 day where the mean squared error was 0.0129.

FCANN: A new approach for extraction and representation of knowledge from ANN trained via Formal Concept Analysis

Nowadays, Artificial Neural Networks (ANN) are being widely used in the representation of different systems and physics processes. Once trained, the networks are capable of dealing with operational conditions not seen during the training process, keeping tolerable errors in their responses. However, humans cannot assimilate the knowledge kept by those networks, since such implicit knowledge is difficult to be extracted. In this work, Formal Concept Analysis (FCA) is being used in order to extract and represent knowledge from previously trained ANN. The new FCANN approach permits to obtain a complete canonical base, non-redundant and with minimum implications, which qualitatively describes the process being studied. The approach proposed has a sequence of steps such as the generation of a synthetic dataset. The variation of data number per parameter and the discretization interval number are adjustment factors to obtain more representative rules without the necessity of retraining the network. The FCANN method is not a classifier itself as other methods for rule extraction; this approach can be used to describe and understand the relationship among the process parameters through implication rules. Comparisons of FCANN with C4.5 and TREPAN algorithms are made to show its features and efficacy. Applications of the FCANN method for real world problems are presented as case studies.

Categorical data clustering

The clustering problem of categorical data resides in choosing the similarity measure.There are several similarity measures from the ones based on simple matching up to the most complex.We raise the issue: is there a similarity measure containing characteristics that are more stable?We compared nine different similarity measures considering three quality measures.We observed that the simplest measure of similarity presented the best results. Inside the clustering problem of categorical data resides the challenge of choosing the most adequate similarity measure. The existing literature presents several similarity measures, starting from the ones based on simple matching up to the most complex ones based on Entropy. The following issue, therefore, is raised: is there a similarity measure containing characteristics which offer more stability and also provides satisfactory results in databases involving categorical variables? To answer this, this work compared nine different similarity measures using the TaxMap clustering mechanism, and in order to evaluate the clustering, four quality measures were considered: NCC, Entropy, Compactness and Silhouette Index. Tests were performed in 15 different databases containing categorical data extracted from public repositories of distinct sizes and contexts. Analyzing the results from the tests, and by means of a pairwise ranking, it was observed that the coefficient of Gower, the simplest similarity measure presented in this work, obtained the best performance overall. It was considered the ideal measure since it provided satisfactory results for the databases considered.

Comparison of Classifiers Efficiency on Missing Values Recovering: Application in a Marketing Database with Massive Missing Data

Missing data in databases are considered to be one of the biggest problems faced on data mining application. This problem can be aggravated when there is massive missing data in the presence of imbalanced databases. Several techniques as samples deletion, values imputation, values prediction through classifiers and approximation of patterns have been proposed and compared, but these comparisons do not consider adverse conditions found in real databases. In this work, it is presented a comparison of techniques used to classify records from a real imbalanced database with massive missing data, where the main objective is the database pre-processing to recover and select records completely filled for further techniques application. It was compared with other algorithms such as clustering, decision tree, artificial neural networks and Bayesian classifier, expressing their efficiency through ROC curves. Through the results, it can be verified that the problem characterization and database understanding are essential steps for a correct techniques comparison in a real problem. It was observed that artificial neural networks are an interesting alternative for this kind of problem since it was capable to obtain satisfactory results even when dealing with real-world problems.

An artificial neural network structure able to obstacle avoidance behavior used in mobile robots

This article presents an artificial neural network (ANN) structure applied to control a mobile robot movement in dynamically changing environments (environments with mobile obstacles). The proposed structure is a backward neural one. So, it is based on past and future positions, and on a optimal pre-established path. The past positions provide the ANN with memory of the mobile robot previous positions. On the other hand, the future positions provide the ANN with a goal, i.e., where the robot should go. Based on this information, the robot do not lose its goal, even if it has to avoid an obstacle. The results show the efficiency of the ANN in a form of simulations.

A model-based predictive control scheme for steal rolling mills using neural networks

A capital issue in roll-gap control for rolling mill plants is the difficulty to measure the output thickness without including time delays in the control loop. Time delays are a consequence of the possible locations for the output thickness sensor, which usually is located some distance away from the roll gap. In this work, a new model-based predictive control law is proposed. The new scheme is a neural network based predictive control structure which is applied to roll-gap control with outstanding results. It is shown that the neural network based predictive control permits to overcome the existing time delays in the system dynamics. The proposed scheme implements a virtual thickness sensor, which releases an accurate estimate of the actual output thickness. It is shown that the dynamic response of the rolling mill system can be substantially improved by using the proposed controller. Simulation results are presented to illustrate the controller performance.

Using the NextClosure algorithm to extract rules from trained neural networks application in solar energy systems

Due to their capability of dealing with nonlinear problems, artificial neural networks (ANN) is widely used with several purposes. Once trained, they are capable to solve unprecedented situations, keeping tolerable errors in their outputs. However, humans cannot assimilate the knowledge kept by those nets, since such knowledge is implicitly represented by their connections weights. So, in order to facilitate the extraction of rules that describe the knowledge of ANN, formal concept analysis (FCA) and the NextClosure algorithm have been used. Such method is presented in this work, combining ANN, FCA and the NextClosure algorithm to compute the minimal implication base (Stem Base). As an example, solar energy systems are the domain application considered here, due to their importance as substitutes of traditional energy systems.

Parallel and distributed kmeans to identify the translation initiation site of proteins

Prediction of the translation initiation site is of vital importance in bioinformatics since through this process it is possible to understand the organic formation and metabolic behavior of living organisms. Sequential algorithms are not always a viable solution due to the fact that mRNA databases are normally very large, resulting in long processing times. Applying parallel and distributed computing resources to such databases could help reduce this time. The objective of this article is to present a class balancing solution for the translation initiation site process using parallel and distributed computing resources in a hybrid model. The results reveal a speedup of up to 23 times compared to sequential methods and performance rates for accuracy, precision, sensitivity, specificity and adjusted accuracy of 91.15%, 39.83%, 89.11%, 88.93% and 89.02%, respectively, for the Homo sapiens database. For the Drosophila melanogaster database, the speedup was 18.33 times and accuracy, precision, sensitivity, specificity and adjusted accuracy were 95.22%, 43.01%, 90.83%, 90.47% and 90.64%, respectively. Both sets of results are considered important. Thus, the solution presented in this article demonstrated itself viable for the problem in question.

Techniques for Missing Value Recovering in Imbalanced Databases: Application in a Marketing Database with Massive Missing Data

Missing data in databases are considered to be one of the biggest problems faced on data mining application. This problem can be aggravated when there is massive missing data in the presence of imbalanced databases. Several techniques as imputation, classifiers and approximation of patterns have been proposed and compared, but these comparisons do not consider adverse conditions found in real databases. In this work, it is presented a comparison of techniques used to classify records from a real imbalanced database with massive missing data, where the main objective is the database pre-processing to recover and select records completely filled for a further application of the techniques. It was compared algorithms such as clustering, decision tree, artificial neural networks and Bayesian classifier. Through the results, it can be verified that the problem characterization and database understanding are essential steps for a correct techniques comparison in a real problem. It was observed that artificial neural networks are an interesting alternative for this kind of problem since it is capable to obtain satisfactory results even when dealing with real-world problems.