A RNN-LSTM based Predictive Modelling Framework for Stock Market Prediction using Technical Indicators

The successful prediction of the stocks’ future price would produce substantial profit to the investor. In this paper, we propose a framework with the help of various technical indicators of the stock market to predict the future prices of the stock using Recurrent Neural Network based Long Short-Term Memory (LSTM) algorithm. The historical transactional data set is amalgamated with the technical indicators to create a more effective input dataset. The historical data is taken from 2010-2019 ten years in total. The dataset is divided into 80% training set and 20% test set. The experiment is carried out in two phases first without the technical indicators and after adding technical indicators. In the experimental setup, it has been observed the LSTM with technical indicators have significantly reduced the error value by 2.42% and improved the overall performance of the system as compared to other machine learning frameworks that are not accounting the effect of technical indicators.

Feature Engineering Techniques to improve Identification Accuracy for Offline Signature Case-Bases

Handwritten signatures have been widely acclaimed for personal identification viability in educated human society. But, the astronomical growth of population in recent years warrant developing mechanized systems to remove the tedium and bias associated with manual checking. Here the proposed system, performing identification with Nearest Neighbor matching between offline signature images collected temporally. The raw images and their extracted features are preserved using Case Based Reasoning and Feature Engineering principles. Image patterns are captured through standard global and local features, along with some profitable indigenously developed features. Outlier feature values, on detection, are automatically replaced by their nearest statistically determined limit values. Search space reduction possibilities within the case base are probed on a few selected key features, applying Hierarchical clustering and Dendogram representation. Signature identification accuracy is found promising when compared with other machine learning techniques and a few existing well known approaches.

Feature Engineering Techniques to Improve Identification Accuracy for Offline Signature Case-Bases

Handwritten signatures have been widely acclaimed for personal identification viability in educated human society. But, the astronomical growth of population in recent years warrant developing mechanized systems to remove the tedium and bias associated with manual checking. Here the proposed system, performing identification with Nearest Neighbor matching between offline signature images collected temporally. The raw images and their extracted features are preserved using Case Based Reasoning and Feature Engineering principles. Image patterns are captured through standard global and local features, along with some profitable indigenously developed features. Outlier feature values, on detection, are automatically replaced by their nearest statistically determined limit values. Search space reduction possibilities within the case base are probed on a few selected key features, applying Hierarchical clustering and Dendogram representation. Signature identification accuracy is found promising when compared with other machine learning techniques and a few existing well known approaches.

A Novel Approach to Enhance Image Security using Hyperchaos with Elliptic Curve Cryptography

Information security dominate the world. All the time we connect to the internet for social media, banking, and online shopping through various applications. Our priceless data may be hacked by attackers. There is a necessity for a better encryption method to enhance information security. The distinctive features of Elliptic Curve Cryptography (ECC) in particular the key atomity, speedy ciphering and preserving bandwidth captivating its use in multimedia encipher. An encryption method is proposed by incorporating ECC, Secure Hash Algorithm – 256 (SHA-256), Arnold transform, and hyperchaos. Randomly generated salt values are concatenated with each pixel of an image. SHA-256 hash is imposed which produces a hash value of 32-bit, later used to generate the key in ECC. Stronger ciphering is done by applying Arnold’s transformation and hyperchaos thereby achieved more randomness in image. Simulation outcomes and analysis show that the proposed approach provides more confidentiality for color images.

Reasoning on vague ontologies using rough set theory

Web ontologies can contain vague concepts, which means the knowledge about them is imprecise and then query answering will not possible due to the open world assumption. A concept description can be very exact (crisp concept) or exact (fuzzy concept) if its knowledge is complete, otherwise it is inexact (vague concept) if its knowledge is incomplete. In this paper, we propose a method based on the rough set theory for reasoning on vague ontologies. With this method, the detection of vague concepts will insert into the original ontology new rough vague concepts where their description is defined on approximation spaces to be used by extended Tableau algorithm for automatic reasoning. The extended Tableau algorithm by this rough set-based vagueness is intended to answer queries even with the presence of incomplete information.

Parallel and Distributed Pattern Mining

The treatment of large data is difficult and it looks like the arrival of the framework MapReduce is a solution of this problem. This framework can be used to analyze and process vast amounts of data. This happens by distributing the computational work across a cluster of virtual servers running in a cloud or a large set of machines. Process mining provides an important bridge between data mining and business process analysis. Its techniques allow for extracting information from event logs. Generally, there are two steps in process mining, correlation definition or discovery and the inference or composition. First of all, their work mines small patterns from log traces. Those patterns are the representation of the traces execution from a log file of a business process. In this step, the authors use existing techniques. The patterns are represented by finite state automaton or their regular expression; and the final model is the combination of only two types of different patterns whom are represented by the regular expressions (ab)* and (ab*c)*. Second, they compute these patterns in parallel, and then combine those small patterns using the Hadoop framework. They have two steps; the first is the Map Step through which they mine patterns from execution traces, and the second one is the combination of these small patterns as a reduce step. The results show that their approach is scalable, general and precise. It minimizes the execution time by the use of the Hadoop framework.

Analyzing Evolution Patterns of Object-Oriented Metrics

Software evolution is mandatory to keep it useful and functional. However, the quality of the evolving software may degrade due to improper incorporation of changes. Quality can be monitored by analyzing the trends of software metrics extracted from source code as these metrics represent the structural characteristics of a software such as size, coupling, inheritance etc. An analysis of these metric trends will give insight to software practitioners regarding effects of software evolution on its internal structure. Thus, this study analyzes the trends of 14 object-oriented (OO) metrics in a widely used mobile operating system software, Android. The study groups the OO metrics into four dimensions and analyzes the trends of these metrics over five versions of Android software (4.0.2-4.3.1). The results of the study indicate certain interesting patterns for the evaluated dimensions, which can be helpful to software practitioners for outlining specific maintenance decisions to improve software quality.

The Effects of Sampling Methods on Machine Learning Models for Predicting Long-term Length of Stay

The ability to predict the patients with long-term length of stay (LOS) can aid a hospital's admission management, maintain effective resource utilization and provide a high quality of inpatient care. Hospital discharge data from the Rhode Island Department of Health from the time period between 2010 to 2013 reveals that inpatients with long-term stays, i.e. two weeks or more, costs about six times more than those with short stays while only accounting for 4.7% of the inpatients. With the imbalance in the distribution of long-stay patients and short-stay patients, predicting long-term LOS patients becomes an imbalanced classification problem. Sampling methods—balancing the data before fitting it to a traditional classification model—offer a simple approach to the problem. In this work, the authors propose a new resampling method called RUBIES which provides superior predictive ability when compared to other commonly used sampling techniques.

An Arabic Dialects Dictionary Using Word Embeddings

The dialectical Arabic and the Modern Standard Arabic lacks sufficient standardized language resources to enable the tasks of Arabic language processing, despite it being an active research area. This work addresses this issue by firstly highlighting the steps and the issues related to building a multi Arabic dialect corpus using web data from blogs and social media platforms (i.e. Facebook, Twitter, etc.). This is to create a vectorized dictionary for the crawled data using the word Embeddings. In other terms, the goal of this article is to build an updated multi-dialect data set, and then, to extract an annotated corpus from it.

Impact of PDS Based kNN Classifiers on Kyoto Dataset

This article compares the performance of different Partial Distance Search-based (PDS) kNN classifiers on a benchmark Kyoto 2006+ dataset for Network Intrusion Detection Systems (NIDS). These PDS classifiers are named based on features indexing. They are: i) Simple PDS kNN, the features are not indexed (SPDS), ii) Variance indexing based kNN (VIPDS), the features are indexed by the variance of the features, and iii) Correlation coefficient indexing-based kNN (CIPDS), the features are indexed by the correlation coefficient of the features with a class label. For comparative study between these classifiers, the computational time and accuracy are considered performance measures. After the experimental study, it is observed that the CIPDS gives better performance in terms of computational time whereas VIPDS shows better accuracy, but not much significant difference when compared with CIPDS. The study suggests to adopt CIPDS when class labels were available without any ambiguity, otherwise it suggested the adoption of VIPDS.