Linkon Chowdhury

Students dropout prediction for intelligent system from tertiary level in developing country

Students dropout prediction is an indispensable for numerous intelligent systems to measure the national and international loss for developing countries as well as for developed country throughout the world. The main purpose of this research is to develop a dynamic dropout prediction model for universities, institutes and colleges. In this work, We first apply chi square test to separate factors such as gender, financial condition and dropping year to classify the successful from unsuccessful students. The main purpose of applying it is feature selection to data. Degree of freedom is used to P-value (Probability value) for best predicators of dependent variable. After being separation of factors we have had examined by using data mining techniques Classification and Regression Tree (CART) and CHAID tree. Among classification tree, growing methods Classification and Regression Tree (CART) was the most successful in growing the tree with an overall percentage of correct classification than CHAID tree. A maximum tree depth has been reached: 3 levels for the CHAID tree and 4 levels for the CART tree. After generating, the classification matrix rules for CART and CHAID tree (Study outcome) have generated. Both the risk estimated by the cross-validation and the gain diagram suggests that all trees based only enrolment data are not quite good in separating successful from unsuccessful students. Here we have considered most important factors to classify the successful students over unsuccessful students are gender, financial condition and dropping year as well as age, gender, ethnicity, education, work status, and disability and study environment that may in-flounce persistence or dropout of students at university level.

MetaG: a graph-based metagenomic gene analysis for big DNA data

Microbial interactions and relationships are significant for animals, insects and plants. Metagenomic research enables properassessments and analysis for microbial organs and communities. The analysis helps to gain detailed insights on miscopies insects. Recent machine learning techniques focused on algorithms and data mining tools to check the depth of interactions and relationships on metagenomic dataset. Accurate analysis over large genes helps to solve real-world problems for public interest. In this regard, graph-centric big gene dataset representations are very important. De Bruijn graph is one the pivotal media to demonstrate the relationships and interactions of large genes dataset or metagenomic dataset. In this research, mapping-based metagenomic graphical (MetaG) genomes representation has been demonstrated. Data cleaning is done before applying graphical illustration. Random mapping is used to assess the variations in dataset. Euler path-based De Bruijn graph is used to sketch the gene annotation, translations, signaling and coding. This research helps in computational biology to map the genomic information in graphical ways with clear conceptions. Adequate experimental comparisons as well as analysis established the claims with tables and graphs.

MSuPDA: A memory efficient algorithm for sequence alignment

AbstractSpace complexity is a million dollar question in DNA sequence alignments. In this regard, memory saving under pushdown automata can help to reduce the occupied spaces in computer memory. Our proposed process is that anchor seed (AS) will be selected from given data set of nucleotide base pairs for local sequence alignment. Quick splitting techniques will separate the AS from all the DNA genome segments. Selected AS will be placed to pushdown automata’s (PDA) input unit. Whole DNA genome segments will be placed into PDA’s stack. AS from input unit will be matched with the DNA genome segments from stack of PDA. Match, mismatch and indel of nucleotides will be popped from the stack under the control unit of pushdown automata. During the POP operation on stack, it will free the memory cell occupied by the nucleotide base pair.Space complexity is a million dollar question in DNA sequence alignments. In this regard, memory saving under pushdown automata can help to reduce the occupied spaces in computer memory. Our proposed process is that anchor seed (AS) will be selected from given data set of nucleotide base pairs for local sequence alignment. Quick splitting techniques will separate the AS from all the DNA genome segments. Selected AS will be placed to pushdown automatas (PDA) input unit. Whole DNA genome segments will be placed into PDAs stack. AS from input unit will be matched with the DNA genome segments from stack of PDA. Match, mismatch and indel of nucleotides will be popped from the stack under the control unit of pushdown automata. During the POP operation on stack, it will free the memory cell occupied by the nucleotide base pair.

Fuzzy Logic over Ontological Annotation and Classification for Spatial Feature Extraction

With the advent of new web technology, Image Annotation and Classification has paved the way for invoking an efficient and effective research area as it is of immense importance in searching images from different categories of relevant images using keywords. This may be an impressive tool in describing image content as object or textual information to classify images. To serve this purpose, many techniques have been lunched for automatic image annotation and classification based on content and exit metadata. Automatic image annotation however, is highly difficult and challengeable. So users have to follow the annotation manually. In this paper, we applied fuzzy logic implication and fuzzy set operation for Historical image classification. We have compared the outcome of how fuzzy classification is better ontological image classification. Fuzzy logic plays an important rule so that the margin of the classification becomes more accurate. Here we imposed fuzzy matrix optimization for Spatial Image classification of Historical image data. Fuzzy matrix determines the optimal values of spatial data which are near about correct with less uncertainty. Fuzzy membership function also works to estimate the values before using in fuzzy matrix. We also pro- posed a manual method for image annotation based on IPTC metadata with a view to retrieving images with its corresponding information for automatic semantic ontological and fuzzy classification using linked data. We strived to experiment on about 400 images of different historical heritages.

Machine-Learning Approach for Ribonucleic Acid Primary and Secondary Structure Prediction from Images

Abstract Ribonucleic acid (RNA) primary and secondary structure prediction and analysis are of utmost importance for assessing the biological organ functionalities. Several RNA structure prediction methods use pattern recognition and artificial intelligence approaches. However, several machine-learning approaches, such as Hidden Markov model, are more interpretable and accurate for an improved structured analysis compared to the artificial intelligence approaches. They employ graphical data analysis to enhance the prediction accuracy. This chapter proposes an RNA prediction approach from RNA images based on hidden Markov model and Chapman Kolmogorov equation with filtering process. Initially, Gaussian filter, box filter, and median filter are applied during the filtering stage. Furthermore, Otsus technique is applied to convert the RNA image into binary image as well as binary matrix. Rank transformation as well as Box and cox transformations are used for binary matrix optimization. In order to classify the RNA structures, Flood fill and Warshall are employed for counting. Finally, Hidden Markov model and Chapman Kolmogorov equations are applied on the classified secondary and tertiary structures of RNA structure prediction.

Pseudoknots Prediction on RNA Secondary Structure Using Term Rewriting

The presences of Pseudoknots generate computational complexities during RNA (Ribonucleic Acid) secondary structure analysis. It is a well known NP hard problem in computational system. It is very essential to have an automated algorithm based system to predict the Pseudoknots from billions of data set. RNA plays a vital role in meditation of cellular information transfer from genes to functional proteins. Pseudoknots are seldom repeated forms that produce misleading computational cost and memory. Memory reducing under bloom filter proposes a memory efficient algorithm for prediction Pseudoknot of RNA secondary structure. RNA Pseudoknot structure prediction based on bloom filter rather than dynamic programming and context free grammar. At first, Structure Rewriting (SR) technique is used to represent secondary structure. Secondary structure is represented in dot bracket representation. Represented secondary structure is separated into two portions to reduce structural complexity. Dot bracket is placed into bloom filter for finding Pseudoknot. In bloom filter, hashing table is used to occupy the RNA based nucleotide. Our proposed algorithm experiences on 105 Pseudoknots in pseudobase and achieves accuracy 66.159% to determine structure.

LiCo: A supervised method for measurement of DNA heterogeneity

In the context of clinical decision support system it is an important task to measure gene similarity effectively for comparative effectiveness studies. It helps clinicians to assess the likely outcomes resulting from their decisions and actions by enabling the capture of past experience as manifested in the collective longitudinal DNA records of genes. But research works about the measurement of gene similarity of heterogeneous DNA records are still few. In order to determine gene similarity of heterogeneous DNA records, we propose a supervised method LiCo combined with three methods namely-Locally Supervised Metric Learning (LSML), interactive Metric learning (iMet) and Composite Distance Integration (Comdi). Our goal is to devise a clinically relevant distance metric to measure gene similarity of heterogeneous DNA records. LSML method generalizes a Mahalanobis distance which is classified the genes of heterogeneous DNA sequences. Interactive Metric learning (iMet) updates the existing metric of genes in heterogeneous DNAs including new records. Finally, Comdi combines multiple similarities from multiple heterogeneous DNA records. We have found accurate solution of large scale DNA records and trims time complexity.

Knowledge base representation for HIV victims in the world

From the birth of the world a large number people have been suffering fatal life killing disease AIDS for various reason. Actually the problem belongs to the people by the people due to the lack of awareness. One the major problem for HIV victims is that it kills their life span by time being unfortunately. In this research, we focus on linked data of semantic web and to achieve the knowledgebase construction for HIV victims. It is obvious that HIV victims are also part of human being and they have full right to get all the support and facilities from all other human. Earlier we have had put concentration on design the linked data for them. We improve our previous work to achieve efficiency on knowledgebase representation. Manipulating our collected data in a structured way by XML parsing on JAVA platform. Our proposed system generates n-triple by considering parsed data. We proceed on an ontology is constructed by Protege which containing information about names, places, awards. A straightforward approach of this work to make the knowledgebase representation of HIV victims more reliable on the web. Our experiments show the effectiveness of knowledgebase construction. Complete knowledgebase construction of HIV victims show the efficient output. The complete knowledgebase construction helps to integrate the raw data in a structured way. The outcome of our proposed system contains the complete knowledgebase. Our experimental results show the strength of our system by retrieving information from ontology in reliable way.

New Dropout Prediction for Intelligent System

The main purpose of this research is to develop a dynamic dropout prediction model for universities, institutes and colleges. In this work, we first identify dependent and independent variables and dropping year to classify the successful from unsuccessful students. Then we have classify the data using Support Vector Machines(SVM). SVM helped the data set to be properly design and manipulated . The main purpose of applying this identification is to design a Knowledge Base which is sometimes known as joint probability distribution . The concepts of propositional logic helped to build the knowledge Base. Bayes theorem will perform the prediction by collecting the information from knowledge Base. Here we have considered most important factors to classify the successful students over unsuccessful students are gender, financial condition and dropping year. We also consider the socio-demographic variables such as age, gender, ethnicity, education, work status, and disability and study environment that may in-flounce persistence or dropout of students at university level