Ahmed Sharaf Eldin

Mining arabic text using soft-matching association rules

Text mining concerns the discovery of knowledge from unstructured textual data. One important task is the discovery of rules that relate specific words and phrases. Textual entries in many database fields exhibit minor variations that may prevent mining algorithms from discovering important patterns. Variations can arise from typographical errors, misspellings, abbreviations, as well as other sources like ambiguity. Ambiguity may be due to the derivation feature, which is very common in the Arabic language. This paper introduces a new system developed to discover soft-matching association rules using a similarity measurements based on the derivation feature of the Arabic language. In addition, it presents the features of using Frequent Closed Item-sets (FCI) concept in mining the association rules rather than Frequent Itemsets (FI).

Agile Tailoring Tool (ATT): A Project Specific Agile Method

after decades from introducing and using agile methodologies, project mangers realized that no methodology is sufficient by itself. Thus, merging their principles is the solution yet no formal solution has been proposed. Relying on previous work, ATT provides a mathematical model to act as a tailoring tool to formulate a new agile method based on experienced agile methods and the project specifications. It requires project managers to understand well the project requirements in terms of SDLC phases, and accordingly the new agile methodology is tailored.

Enhancing packet forwarding in Mobile Ad hoc Networks by exploiting the Information Dispersal Algorithm

A Mobile Ad hoc Network (MANET) is a transitory, dynamic, decentralised, self-organised and infrastructure-less network. This unique nature introduced new challenges in the security research field such that it became imperative to design new security mechanisms that suit these characteristics. The proposed work contributes to the problem of securing data forwarding in the network layer; this is accomplished through exploiting the redundancy of multi-path routing for forwarding data packets. To maximise the possibility of successful delivery to a destination, each node generates a table with a reliability index for each of its neighbour without performing neighbourspsila monitoring; based on that reliability value, sending/receiving traffic is prioritised. This work evaluates the effectiveness of REliable and Efficient Forwarding (REEF) using the Information Dispersal Algorithm (IDA), REEF-IDA; the conducted simulation shows that REEF-IDA outperforms REEF in which it increases the throughput of the network while it decreases both end-to-end delays and packet loss ratio.

Bibliometric analysis of Egyptian publications on Hepatitis C virus from PubMed using data mining of an in-house developed database (HCVDBegy)

Bibliometric analysis of Egyptian literature on HCV provides the intelligence needed for decision makers and gives an insight into research productivity in this area. We propose our database (HCVDBegy) on MS-SQL server by querying PubMed for “HCV and Egypt” with time limit till 31st March 2013. Fifty eight out of the 716 records were excluded and the rest 658 were divided into 22 domains. Analysis used data mining add-ins for Microsoft Excel, including association and regression algorithms. A fluctuation in numbers of papers was noticed from 2004 to 2009 with a steady increase onward. Eighty six percent of publications were the contribution of three or more authors. Top publishing bodies were Cairo and Ain Shams Universities, Faculty of Medicine, National Research Center and National Cancer Institute. Three Egyptian journals came on top, whereas other publishing journals were mainly from the USA. Few controlled clinical trials and meta-analyses were published. HCV epidemiology, review articles and sequence analysis domains were the most cited. Forecasting model showed a breakthrough in numbers of publications on 2013 and 2014 than those forecasted. Dependency network based on association rule model of MeSH topics was also extensively analyzed. Number of publications showed a promising increase which points to the better national awareness of HCV problem. Studying MeSH terms clustering showed some hot topics. We recommend that the PubMed should alarm authors of the challenges of author affiliations. HCVDBegy availability opens the door for more drill down analysis for decision makers.

Response Prediction for Chronic HCV Genotype 4 Patients to DAAs

Hepatitis C virus (HCV) is a major cause of chronic liver disease, end stage liver disease and liver cancer in Egypt. Genotype 4 is the prevalent genotype in Egypt and has recently spread to Southern Europe particularly France, Italy, Greece and Spain. Recently, new direct acting antivirals (DAAs) have caused a revolution in HCV therapy with response rates approaching 100%. Despite the diversity of DAAs, treatment of chronic hepatitis C genotype 4 has not yet been optimized. The aim of this study is to build a framework to predict the response of chronic HCV genotype 4 patients to various DAAs by applying Data Mining Techniques (DMT) on clinical information. The framework consists of three phases which are data preprocessing phase to prepare the data before applying the DMT; DM phase to apply DMT, evaluation phase to evaluate the performance and accuracy of the built prediction model using a data mining evaluation technique. The experimental results showed that the model obtained acceptable results.

ConteSaG: context-based keyword search over multiple heterogeneous graph-modeled data

Following the NoSQL (Not Only SQL) movement, more research work and applications are looking towards graph databases for their dynamic schema and natural representation of complex data1. In order to access/ search graph data in a single source, a number of search methods have been proposed in the literature. These methods range from graph query languages, pattern queries, template/form based search, to keyword search. One key challenge of these methods is the expressiveness and ease of use trade-off for query formulation. Formulating queries becomes more challenging when querying multiple heterogeneous data sources and when users are unaware of the structure of the underlying data. This paper reviews various methods proposed in the literature to query graph modeled data in two different settings; namely, single source and multiple heterogeneous data sources. Furthermore, the paper proposes ConteSaG, a technique for transparently querying multiple heterogeneous data sources. ConteSaG employs graph database to represent data residing in local sources with no need to create complex global schema or even to upfront integrate all data in a central source. Moreover, ConteSaG provides a context-based keyword search over the graph representations. Context-based keyword search allows users to search multiple data sources by determining the context of search terms without the need to have complete knowledge about the structure of data in the local sources or writing queries in a specific query language.

POSTER: A Behavioral Biometric Authentication Framework on Smartphones

To protect smartphones from unauthorized access, the user has the option to activate authentication mechanisms : PIN, Password, or Pattern. Unfortunately, these mechanisms are vulnerable to shoulder-surfing, smudge and snooping attacks. Even the traditional biometric based systems such as fingerprint or face, also could be bypassed. In order to protect smartphones data against these sort of attacks, we propose a behavioral biometric authentication framework that leverages the users behavioral patterns such as touchscreen actions, keystroke, application used and sensor data to authenticate smartphone users. To evaluate the framework, we conducted a field study in which we instrumented the Android OS and collected data from 52 participants during 30-day period. We present the prototype of our framework and we are working on its components to select the best features set that can be used to build different modalities to authenticate users on different contexts. To this end, we developed only one modality, a gesture authentication modality, which authenticate smartphone users based on touch gesture. We evaluated this authentication modality on about 3 million gesture samples based on two schemes, classification scheme with EER~0.004, and anomaly detection scheme with EER~0.10.

Bimodal behavioral authentication framework based on decision fusion

The majority of proposed behavioral biometric systems on smartphones are unimodal based, which rely only on a single source of information such as gesture or keystrokes. Unfortunately, these systems are suffering from some problems such as noisy data and non-universality. Moreover, they provide lower authentication accuracy in compare with the physiological biometrics such as Face. To address these problems, we developed a bimodal authentication framework based on decision fusion. We conducted a field study by instrumenting the Android OS. We analyzed data from 52 participants during 30-day period. We present the prototype of our framework, where we developed two authentication modalities. First, a gesture authentication modality, which authenticate smartphone users based on touch gesture. Second, a keystrokes authentication modality, which authenticate smartphone users based on the way they type. We evaluated each authentication modality based on two schemes, classification scheme and anomaly detection scheme. Then we used the decision fusion method to enhance the accuracy of detection.

Enhancing Aggregation over Uncertain Databases

Queries with aggregation represent an important aspect in database systems. They are widely used in online analytical processing, decision support systems, and data analytics. Aggregate functions usually perform calculations on a set of values of a particular column and return a single summarized value. However, handling aggregate functions becomes a challenge when dealing with uncertain data as there can be an exponential number of possible instances, with potentially different aggregation results for each one. The aim of this paper is to enhance aggregate queries over uncertain databases through a twofold aspect: First, proposing a Probability-Based Aggregation (PBA) technique that considers the probability of each instance in the database. Second, proposing a Probability-Based Entropy (PBE) technique that introduces a new class of aggregate functions to measure the level of uncertainty over databases. Entropy and information gain are two well-known measures stemmed from the information theory but can be used in uncertain databases. The two measures, if used as two aggregate functions in uncertain databases, will allow for more data analytics and mining. Experimental results show that the proposed aggregation technique (PBA) outperforms other similar techniques in terms of precision, recall, uncertainty density, and answer decisiveness. Moreover, using the proposed probabilistic entropy function (PBE) which considers the probability of each instance while calculating the entropy helps in identifying the threshold that gives the maximum information gain.

A Grammatical Inference Sequential Mining Algorithm for Protein Fold Recognition

Protein fold recognition plays an important role in computational protein analysis since it can determine protein function whose structure is unknown. In this paper, a Classified Sequential Pattern mining technique for Protein Fold Recognition (CSPF) is proposed. CSPF technique consists of two main phases: the sequential mining pattern phase and the fold recognition phase. In the sequential mining pattern phase, Mix & Test algorithm is developed based on Grammatical Inference, which is used as a training phase. Mix & Test algorithm minimizes I/O costs by one database scan, discovers subsequence combinations directly from sequences in memory without searching the whole sequence file, has no database projection, handles gaps, and works with variant length sequences without having to align them. In addition, a parallelized version of Mix & Test algorithm is applied to speed up Mix & Test algorithm performance. In the fold recognition phase, unknown protein folds are predicted via a proposed testing function. To test the performance, 36 SCOP protein folds are used, where the accuracy rate is 75.84% for training data and 59.7% for testing data.