Machine Learning

This research is related to the usefulness of different machine learning methods in forecasting time series on financial markets. The main issue in this field is that economic managers and scientific society are still longing for more accurate forecasting algorithms. Fulfilling this request leads to an increase in forecasting quality and, therefore, more profitability and efficiency. In this paper, while we introduce the most efficient features, we will show how valuable results could be achieved by the use of a financial time series technical variables that exist on the Tehran stock market. The suggested method benefits from regression-based machine learning algorithms with a focus on selecting the leading features to find the best technical variables of the inputs. The mentioned procedures were implemented using machine learning tools using the Python language. The dataset used in this paper was the stock information of two companies from the Tehran Stock Exchange, regarding 2008 to 2018 financial activities. Experimental results show that the selected technical features by the leading methods could find the best and most efficient values for the parameters of the algorithms. The use of those values results in forecasting with a minimum error rate for stock data.

Stock price prediction is one of the crucial concepts in finance area. Machine learning can provide the opportunity for traders and investors to predict stock prices more accurately. In this paper, Closing Price is dependent variable and First Price, Last Price, Opening Price, Today’s High, Today’s Low, Volume, Total Index of Tehran Stock Exchange, Brent Index, WTI Index and Exchange Rate are independent variables. Seven different machine learning algorithms are implemented to predict stock prices. Those include Bayesian Linear, Boosted Tree, Decision Forest, Neural Network, Support Vector, and Ensemble Regression. The sample of the study is fifteen oil and gas companies active in the Tehran Stock Exchange. For each stock the data from the September 23, 2017 to September 23, 2019 gathered. Each algorithm provided two metrics for performance: Root Mean Square Error and Mean Absolute Error. By comparing the aforementioned metrics, the Bayesian Linear Regression had the best performance to predict stock price in the oil and gas industry in the Tehran Stock Exchange.

Successful trades in financial markets have to be conducted close to the key recurrent points. Researchers have recently developed diverse systems to help the identification of these points. Technical analysis is one of the most valid and all-purpose kinds of these systems. With its numerous rules, the technical analysis endeavors to create well-timed and correct signals so that these points are identified. However, one of the drawbacks of this system is its overdependence on human analysis and knowledge in selecting and applying these rules. Employing the three tools of genetic algorithm, fuzzy logic, and neural network, this study attempts to develop an intelligent trading system based on the recognized rules of the technical analysis. Indeed, the genetic algorithm will assist with the optimization of technical rules owing to computing complexities. The fuzzy inference will also help the recognition of the total current condition in the market. It is because a set of rules will be selected based on the market kind (trending or non-trending). Finally, the signal developed by every rule will be translated into a single result (buy, sell, or hold). The obtained results reveal that there is a statistically meaningful difference between a stock's buy and hold and the trading system proposed by this research. In other words, our proposed system displays an extremely higher profitability potential.

Businesses are increasingly interested in how big data, artificial intelligence, machine learning, and predictive analytics can be used to increase revenue, lower costs, and improve their business processes. In this paper, we describe how we have developed a data-driven machine learning method to optimize the collection process for a debt collection agency. Precisely speaking, we create a framework for the data-driven scheduling of outbound calls made by debt collectors. These phone calls are used to persuade debtors to settle their debt, or to negotiate payment arrangements in case debtors are willing, but unable to repay. We determine daily which debtors should be called to maximize the amount of delinquent debt recovered in the long term, under the constraint that only a limited number of phone calls can be made each day. Our approach is to formulate a Markov decision process and, given its intractability, approximate the value function based on historical data through the use of state-of-the-art machine learning techniques. Precisely, we predict the likelihood with which a debtor in a particular state is going to settle its debt and use this as a proxy for the value function. Based on this value function approximation, we compute for each debtor the marginal value of making a call. This leads to a particularly straightforward optimization procedure, namely, we prioritize the debtors that have the highest marginal value per phone call. We believe that our optimized policy substantially outperforms the current scheduling policy that has been used in business practice for many years. Most importantly, our policy collects more debt in less time, whilst using substantially fewer resources leading to a large increase in the amount of debt collected per phone call.

Online reviews have become essential aspect in E-commerce platforms due to its role for assisting customers’ buying choices. Furthermore, the most helpful reviews that have some attributes are support customers buying decision; therefore, there is needs for investigating what are the attributes that increase the Review Helpfulness (RH). This research paper proposed novel model called inclusive review helpfulnessmodel (IRHM) can be used to detect the most attributes affecting the RH and build classifier that can predict RH based on these attributes. IRHM is implemented on Amazon.com using collection of reviews from different categories. The results show that IRHM can detect the most important attributes and classify the reviews as helpful or not with accuracy of 94%, precision of 0.20 and had excellent area under curve close to 0.94.

Handwriting Analysis has been used for a very long time to analyze an individual’s suitability for a job, and is in recent times, gaining popularity as a valid means of a person’s evaluation. Extensive Research has been done in the field of determining the Personality Traits of a person through handwriting. We intend to analyze an individual’s personality by breaking it down into the Big Five Personality Traits using their handwriting samples. We present a dataset that links personality traits to the handwriting features. We then propose our algorithm - consisting of one ANN based model and PersonaNet, a CNN based model. The paper evaluates our algorithm’s performance with baseline machine learning models on our dataset. Testing our novel architecture on this dataset, we compare our algorithm based on various metrics, and show that our novel algorithm performs better than the baseline Machine Learning models.

Due to enchantment in network technology, the worldwide numbers of internet users are growing rapidly. Most of the internet users are using online purchasing from various sites. Due to new online shopping trends over the internet, the seller needs to predict the online customer’s choice. This field is a new area of research for machine learning researchers. A random forest (RF) machine learning method is a widely used classification method. It is mainly based on an ensemble of a single decision tree. Online e-commerce websites accumulate a massive quantity of data in large dimensions. A Random Forest is an efficient filter in high-dimensional data to reliably classify consumer behaviour factors. This research article mainly proposed an extension of the Random Forest classifier named “Weighted Random Forests” (wRF), which incorporates tree-level weights to provide much more accurate trees throughout the calculation as well as an assessment of vector relevance. The weighted random forest algorithm incorporates the C4.5 method named a “Hybrid Weighted Random Forest” (HWRF) to forecast online consumer purchasing behaviour. The experimental results influence the quality of the proposed method in the prediction of the behaviour of online buying customers over existing methods.

The Special Issue of the Journal of Information Technology Management (JITM) is publishing very selective papers on information management, Internet of Things (IoT), Algorithms, Quality of Service (QoS),Tourists Perception , Technology in higher education, integrated systems, enterprise management, Self-Service Technology (SST) , cultural thoughts, strategic contributions, management information systems, and cloud computing. We received numerous papers for this special issue but after an extensive peer-review process, eight papers were finally selected for publication. In the digital age, the management of electronic archives became a trend as well as the focus of management development in many institutions.

Sentiment Analysis or opinion mining is NLP's method to computationally identify and categorize user opinions expressed in textual data.  Mainly it is used to determine the user's opinions, emotions, appraisals, or judgments towards a specific event, topic, product, etc. is positive, negative, or neutral. In this approach, a huge amount of digital data generated online from blogs and social media websites is gathered and analyzed to discover the insights and help make business decisions. Social media is web-based applications that are designed and developed to allow people to share digital content in real-time quickly and efficiently.  Many people define social media as apps on their Smartphone or tablet, but the truth is, this communication tool started with computers. It became an essential and inseparable part of human life. Most business uses social media to market products, promote brands, and connect to current customers and foster new business. Online social media data is pervasive. It allows people to post their opinions and sentiments about products, events, and other people in the form of short text messages. For example, Twitter is an online social networking service where users post and interact with short messages, called "tweets." Hence, currently, social media has become a prospective source for businesses to discover people's sentiments and opinions about a particular event or product. This paper focuses on the development of a Multinomial Naïve Bayes Based social media data emotion analyzer and sentiment classifier. This paper also explains various enriched methods used in pre-processing techniques. This paper also focuses on various Machine Learning Techniques and steps to use the text classifier and different types of language models.

As electronic commerce has gained widespread popularity, payments made for users' transactions through credit cards also gained an equal amount of reputation. Whenever shopping through the web is made, the chance for the occurrence of fraudulent activities are escalating. In this paper, we have proposed a three-phase scheme to detect fraudulent activities. A profile for the card users based on their behavior is created by employing a machine learning technique in the second phase extraction of a precise communicative pattern for the card users depending upon the accumulated transactions and the user's earlier transactions. A collection of classifiers are then trained based on all behavioral pattern. The trained collection of classifiers are then used to detect the fraudulent online activities that occurred. If an emerging transaction is fraudulent, feedback is taken, which resolves the drift's difficulty in the notion. Experiments performed indicated that the proposed scheme works better than other schemes.

Learning effective segment depictions and resemblance measures are fundamental to the recuperation execution of a substance based picture recuperation (CBIR) structure. Regardless of wide research tries for a significant long time, it stays one of the most testing open gives that broadly impedes the achievements of real-world CBIR structures. The key test has been credited to the extraordinary "semantic hole" subject that happens between low-level photo pixels got by technologies and raised close semantic thoughts saw by a human. Among various techniques, AI has been successfully analyzed as a possible course to interface the semantic gap in the whole deal. Impelled by late triumphs of significant learning techniques for PC vision and various applications, in this paper, we try to address an open issue: if significant learning is a longing for spreading over the semantic gap in CBIR and how much updates in CBIR endeavors can be cultivated by exploring the front line significant learning methodology for learning feature depictions and likeness measures. Specifically, we explore a structure of significant learning with application to CBIR assignments with a wide game plan of definite examinations by investigating front line significant learning methodologies for CBIR endeavors under moved settings. From our exploratory examinations, we find some encouraging results and compress some huge bits of information for upcoming research.

Learning effective segment depictions and resemblance measures are fundamental to the recuperation execution of a substance based picture recuperation (CBIR) structure. Regardless of wide research tries for a significant long time, it stays one of the most testing open gives that broadly impedes the achievements of real-world CBIR structures. The key test has been credited to the extraordinary "semantic hole" subject that happens between low-level photo pixels got by technologies and raised close semantic thoughts saw by a human. Among various techniques, AI has been successfully analyzed as a possible course to interface the semantic gap in the whole deal. Impelled by late triumphs of significant learning techniques for PC vision and various applications, in this paper, we try to address an open issue: if significant learning is a longing for spreading over the semantic gap in CBIR and how much updates in CBIR endeavors can be cultivated by exploring the front line significant learning methodology for learning feature depictions and likeness measures. Specifically, we explore a structure of significant learning with application to CBIR assignments with a wide game plan of definite examinations by investigating front line significant learning methodologies for CBIR endeavors under moved settings. From our exploratory examinations, we find some encouraging results and compress some huge bits of information for upcoming research.

This paper describes a series of experiments in using logistic regression machine learning as a method for entity resolution. From these experiments the authors concluded that when a supervised ML algorithm is trained to classify a pair of entity references as linked or not linked pair, the evaluation of the model’s performance should take into account the transitive closure of its pairwise linking decisions, not just the pairwise classifications alone. Part of the problem is that the measures of precision and recall as calculated in data mining classification algorithms such as logistic regression is different from applying these measures to entity resolution (ER) results.. As a classifier, logistic regression precision and recall measure the algorithm’s pairwise decision performance. When applied to ER, precision and recall measure how accurately the set of input references were partitioned into subsets (clusters) referencing the same entity. When applied to datasets containing more than two references, ER is a two-step process. Step One is to classify pairs of records as linked or not linked. Step Two applies transitive closure to these linked pairs to find the maximally connected subsets (clusters) of equivalent references. The precision and recall of the final ER result will generally be different from the precision and recall measures of the pairwise classifier used to power the ER process. The experiments described in the paper were performed using a well-tested set of synthetic customer data for which the correct linking is known. The best F-measure of precision and recall for the final ER result was obtained by substantially increasing the threshold of the logistic regression pairwise classifier.

The quest for improving the software quality has given rise to various studies which focus on the enhancement of the quality of software through various processes. Code smells, which are indicators of the software quality have not been put to an extensive study for as to determine their role in the prediction of defects in the software. This study aims to investigate the role of code smells in prediction of non-faulty classes. We examine the Eclipse software with four versions (3.2, 3.3, 3.6, and 3.7) for metrics and smells. Further, different code smells, derived subjectively through iPlasma, are taken into conjugation and three efficient, but subjective models are developed to detect code smells on each of Random Forest, J48 and SVM machine learning algorithms. This model is then used to detect the absence of defects in the four Eclipse versions. The effect of balanced and unbalanced datasets is also examined for these four versions. The results suggest that the code smells can be a valuable feature in discriminating absence of defects in a software.

This paper presents an approach that uses both genetic algorithm (GA) and fuzzy inference system (FIS), for feature selection for descriptor in a quantitative structure activity relationships (QSAR) classification and prediction problem. Unlike the traditional techniques that employed GA, the FIS is used to evaluate an individual population in the GA process. So, the fitness function is introduced and defined by the error rate of the GA and FIS combination. The proposed approach has been implemented and tested using a data set with experimental value anti-human immunodeficiency virus (HIV) molecules. The statistical parameters q2 (leave many out) is equal 0.59 and r (coefficient of correlation) is equal 0.98. These results reveal the capacity for achieving subset of descriptors, with high predictive capacity as well as the effectiveness and robustness of the proposed approach.

One of the most critical investment issues faced by different investors is choosing an optimal investment portfolio and balancing risk and return in a way that, maximizes investment returns and minimize the investment risk. So far, many methods have been introduced to form a portfolio, the most famous of the Markowitz approach. The Markowitz mean-variance approach is widely known in the world of finance and, it marks the foundation of every portfolio theory. The mean-variance theory has many practical drawbacks due to the difficulty in estimating the expected return and covariance for different asset classes. In this study, we use the Hierarchical Risk Parity (HRP) machine learning technique and compare the results with the three methods of Minimum Variance (MVP), Uniform Distribution (UNIF), and Risk Parity (RP). To conduct this research, the adjusted price of 50 listed companies of the Tehran Stock Exchange for 2018-07-01 to 2020-09-29 has been used. 70% of the data are considered as in-sample and the remaining 30% as out-of-sample. We evaluate the results using four criteria: Sharp, Maximum Drawdown, Calmer, Sortino. The results show that the MVP and, UNIF approach within the in-sample and, the UNIF and HRP approach out-of-sample have the best performance in sharp measure.

E-commerce websites, based on their structural ontology, provides access to a wide range of options and the ability to deal directly with manufacturers to receive cheaper products and services as well as receiving comments and ideas of the users on the provided products and services. This is a valuable source of information, which includes a large number of user reviews. It is difficult to check the bulk of the comments published manually and non-automatically. Hence, sentiment analysis is an automated and relatively new field of study, which extracts and analyzes people's attitudes and emotions from the context of the comments. The primary objective of this research is to analyze the content of users' comments on online sale e-commerce websites of handcraft products. Sentiment analysis techniques were used at sentence level and machine learning approach.  First, the pre-processing steps and TF-IDF method were implemented on the comments text. Next, the comments text were classified into two groups of products and services comments using Support Vector Machine (SVM) algorithm with 99.2% accuracy. Finally, the sentiment of comments was classified into three groups of positive, negative and neutral using XGBoost algorithm. The results showed, 95.23% and 95.12% accuracies for classification of sentiments in comments about products and services, respectively.

With the emergence of the World Wide Web, Electronic Commerce (E-commerce) has been growing rapidly in the past two decades. Intelligent agents play the main role in making the negotiation between different entities automatically.  Automated negotiation allows resolving opponent agents' mutual concerns to reach an agreement without the risk of losing individual profits. However, due to the unknown information about the opponent's strategies, automated negotiation is difficult. The main challenge is how to reveal the optimal information about the opponent's strategy during the negotiation process to propose the best counter-offer. In this paper, we design a buyer agent which can automatically negotiate with the opponent using artificial intelligence techniques and machine learning methods. The proposed buyer agent is designed to learn the opponent's strategies during the negotiation process using four methods: "Bayesian Learning", "Kernel Density Estimation", "Multilayer Perceptron Neural Network", and "Nonlinear Regression". Experimental results show that the use of machine learning methods increases the negotiation efficiency, which is measured and evaluated by parameters such as the rate agreement (RA), average buyer utility (ABU), average seller utility (ASU), average rounds (AR). Rate agreement and average buyer utility have increased from 58% to 74% and 90% to 94%, respectively, and average rounds have decreased from 10% to 0.04%.

Building and maintaining brand loyalty is a vital issue for market research departments. Various means, including online advertising, helps with promoting loyalty to the brand amongst users. The present paper studies intelligent web advertisements with an eye-tracking technique that calculates users’ eye movements, gaze points, and heat maps. This paper examines different features of an online ad and their combinations, such as underlining words and personalization by eye-tracking. These characteristics include underlining, changing color, number of words, personalizing, inserting a related photograph, and changing the size and location of the advertisement on a website. They help advertisers to improve their ability to manage the ads by increasing users' attention. Moreover, the current research argues the impact of gender on users' visual behavior for advertising features in different Cognitive Demand (CD) levels of tasks while avoiding interruption of users’ cognitive processes with eye-tracking techniques. Also, it provides users the most relevant advertisement compatible with CD level of a task by Support Vector Machine (SVM) algorithm with high accuracy. This paper consists of two experiments that one of them has two phases. In the first and second experiments, a news website alongside an advertisement and an advertising website is shown to the users. The results of the first experiment revealed that personalizing and underlining the words of the ad grabs more attention from users in a low CD task. Furthermore, darkening the background promotes users' frequency of attention in a high CD task. By analyzing the impact of gender on users' visual behavior, males are attracted to the advertisement with red-colored words sooner than females during the high CD task. Females pay more prolonged and more frequent attention to the ads with red-colored words and larger sizes in the low CD task. The second experiment shows that the gazing start point of users with a right to left mother tongue language direction is mainly in the middle of the advertising website.

Q&A forums are designed to help users in finding useful information and accessing high-quality content posted by other users in text forums. Automatically identifying high-quality replies posted in response to the initial posts not only provides users with appropriate content, but also saves their time. Existing methods for classifying user replies based on their quality, try to extract quality features from both the textual content and metadata of the replies. This feature engineering step is a time and labor-intensive task. The current study addresses this problem by proposing new model based on deep learning for detecting quality user replies using only raw textual content. Specifically, we propose a long short-term memory (LSTM) model that exploits the embeddings from language models (ELMo) for representing words as contextual numerical vectors. We compared the effectiveness of the proposed model with four traditional machine learning models on the TripAdvisor for New York City (NYC) and the Ubuntu Linux distribution online forums datasets. Experimental results indicated that the proposed model significantly outperformed the four traditional algorithms on both datasets. Moreover, the proposed model achieved about 16% higher accuracy compared to that obtained by the traditional algorithms trained on both textual and quality dimension features.