Search

The recent success of the application of Artificial Intelligence in the financial sector has resulted in more firms relying on stochastic models for predicting the behaviour of the market. Everyday, quantitative analysts strive to attain better accuracies from their machine learning models for forecasting returns from stocks. Support Vector Machine (SVM) and Random Forest based regression models are known for their effectiveness in accurately predicting closing prices. In this work, we propose a technique for analyzing and predicting stock prices of companies using the aforementioned algorithms as an ensemble. Datasets from India's National Stock Exchange (NSE) containing basic market price information are preprocessed to include well known leading technical indicators as features. Feature selection, which ranks features based on their degree of influence on the final closing price has been incorporated to reduce the size of the training dataset. Additionally, we evaluate the effectiveness of considering the public opinion of a company by employing sentiment analysis. Using a trained Word2Vec model, company specific hash-tagged posts from Twitter are classified as positive or negative. Our proposed ensemble model is then trained on a new dataset which combines the technical indicator data along with the aggregated number of positive/negative tweets of a company over time. Our experiments indicate that in some scenarios, the ensemble model performs better than the constituent models and is highly dependent of the nature and size of the training data. However, combining technical indicator data with aggregated positive/negative tweet counts has a negligible effect on the performance of the ensemble model.

Information sharing and review platforms has generated large volumes of opinionated data which is usually in unstructured form. With the help of Sentiment Analysis, this data can be transformed into structured data which can be useful for commercial applications such as product reviews and feedback, marketing analysis, etc. The purpose of this work is to analyzes the performance of three classifiers(SVM, Naive Bayes, and Logistic Regression) with respect to providing positive or negative sentiment for three different scenarios(Movie Reviews, Election Opinions, and Food Reviews). The three classifiers are compared using fixed set of preprocessing steps and four different weighting schemes(Term frequency inverse document frequency (TFIDF), Term frequency inverse class frequency (TFICF), Mutual Information (MI), and X2 statistic (CHI)). The controlled experimental results showed that Logistic Regression classifier performs better in terms of overall accuracy when MI is used as weighting scheme.

Efficient performance of a disk I/O operation involves a multitude of factors such as the type of the disk, I/O scheduling, and the type of the file system used. Due to the various types of file systems available, with each having different structure and logic, properties of speed, flexibility, security, size and more, it becomes imperative to have an objective overview of the merits and demerits of each file system according to the needs of the users. In this work, we present a thorough performance evaluation of ext4, NTFS and Btrfs filesystems along with CFQ, NOOP and Deadline I/O schedulers tested on regular hard disk drives and SSDs.

Convolution Neural Networks (CNN) models have shown remarkable results for classifying text and sentiments. In this paper, we present our approach on the task of classifying movie reviews using the word embedding on a large-scale dataset provided by Stanford. We compare the accuracy of word-based CNN models for different kernel sizes with structured and unstructured sentences. We use the standard GLoVe for word embedding and use MXNet framework to perform our experiments. Our results suggest that word orderings in a sentence do not affect the classification accuracy of CNN models significantly. However, we also note that accuracy may depend on frequency of occurrences of words.

Short-term trading is a difficult task due to fluctuating demand and supply in the stock market. These demands and supply are reflected in stock prices. The stock prices may be predicted using technical indicators. Most of the existing literature considered the limited technical indicators to measure short-term prices. We have considered 33 different combinations of technical indicators to predict the stock prices. The paper has two objectives, first is the technical indicator feature selection and identification of the relevant technical indicators by using Boruta feature selection technique. The second objective is an accurate prediction model for stocks. To predict stock prices we have proposed ANN (Artificial Neural Network) Regression prediction model and model performance is evaluated using metrics is Mean absolute error (MAE) and Root mean square error (RMSE). The experimental results are better than the existing method by decreasing the error rate in the prediction to 12%. We have used the National Stock Exchange, India (NSE) data for the experiment.

Stock price returns forecasting is challenging task for day traders to yield more returns. In the past, most of the literature was focused on machine learning algorithm to predict the stock returns. In this work, the recurrent neural network (RNN) with long short term memory (LSTM) is studied to forecast future stock returns. It has the ability to keep the memory of historical stock returns in order to forecast future stock return output. RNN with LSTM is used to store recent stock information than old related stock information. We have considered a recurrent dropout in RNN layers to avoid overfitting in the model. To accomplish the task we have calculated stock return based on stock closing prices. These stock returns are given as input to the recurrent neural network. The objective function of the prediction model is to minimize the error in the model. To conduct the experiment, data is collected from the National Stock Exchange, India (NSE). The proposed RNN with LSTM model outperforms compared to an feed forward artificial neural network.

We conduct discrimination among four important life distributions — Log-Normal distribution, Weibull distribution, Gamma distribution and Generalized Exponential distribution. In this paper, we have assumed that the data is coming from either of these four distributions. For the purpose of analyzing lifetime data in the field of reliability analysis, these distributions have been extensively used. Even though adequate amount of research work has been done which focuses on discriminating between two or three distributions, enough work has not been done in choosing among more than three distributions. We have used three goodness of fit tests, namely, KS test, AIC test and BIC test to choose the best fitting model. The task of choosing the right model becomes difficult when we have to discriminate among closely fitting models. We have used probability of correct selection to further validate the results of the discrimination process. The probability of correct selection is calculated using asymptotic distributions. For illustration, we have analyzed two real life data sets.

Performance degradation is unavoidable in server systems and this is because of factors suchas shrinkage of system resources, data corruption, and numerical error accumulation. The resource shrinkageleads to the system failure due to the error propagation. Thus the resource prediction is useful to theadministrator of the system so that an accidental outage can be avoided. It has been observed in past thatmost of the failures occur due to the exhaustion of free physical memory, so here free physical memory of aserver consolidation setup is observed. It is also found that most of the studies in this direction were using themeasurement-based approach with time series models for prediction. This paper reviews the effectiveness ofsuch models and it examines whether volatility is present in the data or not. It checks whether Gauss-Markovassumptions about homoscedasticity holds good for the ordinary least square estimators of such models ornot. This paper applies a combination of AutoRegressive Integrated Moving Average - AutoRegressiveConditional Heteroskedastic (ARIMA-ARCH) model to predict resource usage. Experimental resultsdemonstrate that the goodness of fit of the ARIMA-ARCH Model has improved when compared to the linearARIMA model.

This paper uses a case based study – “product sales estimation” on real-time data to help us understand the applicability of linear and non-linear models in machine learning and data mining. A systematic approach has been used here to address the given problem statement of sales estimation for a particular set of products in multiple categories by applying both linear and non-linear machine learning techniques on a data set of selected features from the original data set. Feature selection is a process that reduces the dimensionality of the data set by excluding those features which contribute minimal to the prediction of the dependent variable. The next step in this process is training the model that is done using multiple techniques from linear & non-linear domains, one of the best ones in their respective areas. Data Remodeling has then been done to extract new features from the data set by changing the structure of the dataset & the performance of the models is checked again. Data Remodeling often plays a very crucial and important role in boosting classifier accuracies by changing the properties of the given dataset. We then try to explore and analyze the various reasons due to which one model performs better than the other & hence try and develop an understanding about the applicability of linear & non-linear machine learning models. The target mentioned above being our primary goal, we also aim to find the classifier with the best possible accuracy for product sales estimation in the given scenario.

The paper aims to determine the forest cover type of the dataset containing cartographic attributes evaluated over four wilderness areas of Roosevelt National Forest of Northern Colorado. The cover type data is provided by US Forest service inventory, while Geographic Information System (GIS) was used to derive cartographic attributes like elevation, slope, soil type etc. Dataset was analyzed, pre processed and feature engineering techniques were applied to derive relevant and non-redundant features. A comparative study of various decision tree algorithms namely, CART, C4.5, C5.0 was performed on the dataset. With the new dataset built by applying feature engineering techniques, Random Forest and C5.0 improved the accuracy by 9% compared to the raw dataset.

This paper tries to establish relation between the power usage and software aging. Software aging is the performance degradation of long running software due to shrinking in physical memory, increase in swap read and write rate and increase in CPU utilization. This paper tries to establish the relation between the Software aging and the power usage. Experimental results demonstrate that CPU utilization increases over a period of time, when the work load remains the constant. Linear Regression analysis is used for establishing this trend.

Nowadays virtualization playing a vital role in the cloud technology. Users deploying websites on virtual private servers called instances which is cost effective and scalable. The explosive growth in World Wide Web, users are able to access things with any device at any time and any place. Increasing the access will constitutes the load and stress on the servers. Sometrics are useful to gauge the performance of various architectures. In this analysis we consider three matrices, they are responsiveness, scalability and efficiency. We have defined response time, memory usage and error rate as key factors for responsiveness, scalability and efficiency respectively. In this paper we investigate the performance of two web servers in different scenario, i.e. Apache - process-based webserver and Nginx - asymmetric multi-process event-driven architecture (AMPED). Our research shows that the Nginx server out performs Apache server in terms of responsiveness and scalability, while Apache ensures the efficiency. The results are useful for those who want to host a website on virtual private server (VPS).

Degradation of the performance is the part of any long running software systems. This is due to memory leakage, unreleased file descriptors, round off errors and disk and memory fragmentation. It has been found that the memory leakage is the primary cause of any software performance degradation. In order to predict the software performance degradation, the analysis of the resource usage is essential. Here the free physical memory of a server virtualised system is analysed using time series analysis.

In this paper, we have proposed a novel K-means algorithm with modified Cosine Distance Measure for clustering of large datasets like Wikipedia latest articles and Reuters dataset. We are customizing Cosine Distance Measure for computing similarity between objects for improving cluster quality. Our method will calculate the similarity between objects by Cosine Distance Measure and then try to bring distance more closer by squaring the distance if it is between 0 to 0.5 else increase it. It will result in minimum Intra-cluster and maximizes Inter-cluster distance value. We are measuring cluster quality in term of Inter and Intra-cluster distances, good Feature weighting such as TF-IDF, Cluster Size and Top terms of the clusters. We have compared K-means algorithm by Cosine and modified Cosine Distance measure by setting performance metric such as Inter-cluster and Intra-cluster distances, Cluster size, Execution time etc. Our experimental result shows in minimizing Intra-cluster by 0.016% and maximizing Inter-cluster distance by 0.012%, reducing the cluster size by 1.5% and reducing sequence file size by 4%, that will result in good cluster quality.

This paper uses a case based study – “product sales estimation” on real-time data to understand the applicability of linear and non-linear models. We use a systematic approach to address the given problem statement of sales estimation for a given product by applying both linear and non-linear techniques on a data set of selected features from the original data set. Feature selection is a process that reduces the dimensionality of the data set by eliminating those features which contribute minimal to the prediction of the dependent variable. The next step is training the model which is done using two techniques from linear & non-linear domains, one of the best ones in their respective areas. Data Re-modeling has then been done to extract new features from the data set by changing the structure of the dataset & the performance of the models is checked again. Data Remodeling often plays a crucial role in boosting classifier accuracies by changing the properties of the dataset. We then try to analyze the reasons due to which one model proves to be better than the other & hence try and develop an understanding about the applicability of linear & non-linear models. The target mentioned above being our primary goal, we also aim to find the classifier with the best possible accuracy for product sales estimation in the given scenario.

Virtualization is rapidly gaining acceptance as a fundamental building block in enterprise data centers and is the core component of cloud computing platforms. It is most known for improving efficiency and ease of management. While this technology is meant to enhance the security of computer systems, some recent attacks show that virtual machine technology has much vulnerability and becomes exposed to security threats. In this paper, we focus on Vulnerability Analysis on Virtualized Environment using First Principles Vulnerability Analysis (FPVA) methodology. This paper analyses the security of Interrupt Descriptor Table on Xen hypervisor and presents outcome of analysis. This paper aids other enthusiasts in better analysing the Security aspects of Xen system using their own programs for security vulnerabilities.

We describe SR-POEM, a Galilean test of the weak equivalence principle that is to be conducted during the free fall portion of the flight of a sounding rocket payload. This test of a single pair of substances will have a measurement uncertainty of {\sigma}({\eta}) < 2 10^17 after averaging the results of eight separate drops, each of 120 s duration. The entire payload is inverted between successive drops to cancel potential sources of systematic error. The weak equivalence principle measurement is made with a set of four of the SAO laser gauges, which have achieved an Allan deviation of 0.04 pm for an averaging time of 30 s. We discuss aspects of the current design with an emphasis on those that bear on the accuracy of the determination of {\eta}. The discovery of a violation ({\eta} \neq 0) would have profound implications for physics, astrophysics and cosmology., Comment: 20 pages, 5 figures (color)

In this paper a novel theoretical model is presented in which the dynamics in a knowledge-based value chain is modeled using Petri nets. From the generic scheme of a knowledge-based value chain, the various components are individually modeled. The theory of Petri nets aptly captures the evolution of knowledge in a system and the process is usually highly interactive in nature. The properties and analysis methods of various classes of Petri nets can be conveniently used to check various constraints while designing the system.

SR-POEM, our principle of equivalence measurement on a sounding rocket, will compare the free fall rate of two substances yielding an uncertainty of E-16 in the estimate of \eta. During the past two years, the design concept has matured and we have been working on the required technology, including a laser gauge that is self aligning and able to reach 0.1 pm per root hertz for periods up to 40 s. We describe the status and plans for this project., Comment: Presented at the Fifth Meeting on CPT and Lorentz Symmetry, Bloomington, Indiana, June 28-July 2, 2010