Journal Volume: 67      No.: 1     Year: 2013
S.No Title Abstract Download
1 Diseases and Pests Identification in Maize - A Multilingual Scenario
Author: Sudeep Marwaha, Punam Bedi and V.K. Yadav      Pages: 107-120
Pests and diseases cause major economic losses to the farmers. The estimate loss per annum due to diseases and pests in a country like India touches to billions of Rupees. Most of the time farmers use over dosages of pesticides and fungicides to save their crop and thus cause environmental hazards. The presented expert system is designed to help farmers to identify diseases and insects attacking maize crop which is neither feasible nor practical by conventional system of extension. Traditional expert systems are based on rules and facts whereas the knowledgebase of this expert system is built using ontology ? the latest knowledge representation technique. OwL is the w3C specifications for building ontologies. It is based on XML and Unicode. Moreover, rule-base knowledgebase is not inherently based on Unicode and thus lacks support for internationalization or for regional languages. The system acts as a tool for transferring the site and crop specific knowledge of various domain experts to the farmers. The system is integrated with the Maize Agridaksh. Agridaksh is a tool for developing online expert system of crops. India being a multilingual society with over 16 major languages and most of the farmers across the country understands their local language only. This system is multilingual and at present contains knowledge in English and Hindi languages. Keywords: Ontology, OwL, Diseases, Pests identification, Maize, Protégé, Semantic web, SPARQL.
2 Cashew Kernel Classification Using Machine Learning Approaches
Author: J. Ashok Kumar, P.R. Rao and A.R. Desai      Pages: 121-129
Understanding patterns present in cashew kernel parameters and building a mathematical model for automatic cashew kernel classification (grading) is an important research area. In this study we attempt to understand the associations present in the cashew kernels and find the best supervised learning model based on kernel parameters of export quality whole cashew grades. There are around 25 export quality cashew grades ranging from wholes to bits and pieces. We have taken top 5 whole grades for this study which are considered important in international market. Parametric techniques like correlation, regression and machine learning approaches like decision trees, logistic regression, artificial neural networks and support vector machines have been used to understand patterns and build an efficient classifier for effective classification of cashew kernels. The results reveal a perfect correlation between kernel length and kernel weight (r = 0.9). Linear regression between the kernel weight and length proved to be sufficient model with different predictor variables (R2 upto 0.92). Classification algorithms were evaluated with different sets of input variables with machine vision perspective. Among the different machine learning techniques used for developing a classification model, back propagation model of artificial neural networks proved to be the best with an average classification accuracy of 85%. Keywords: Cashew kernels, Machine learning, Regression, Artificial neural networks.
3 Hindi Supplement
Author: ISAS      Pages: 131-137
4 Preface
Author: ISAS      Pages: 1
5 Predicting Economic Traits in Murrah Buffaloes with Connectionist Models
Author: A.K. Sharma, D.K. Jain, A.K. Chakravarty, R. Malhotra and A.P. Ruhil      Pages: 1-11
In this paper, several predictive models based on connectionist paradigms and conventional multiple regression approach are proposed to predict milk yield in different lactations as well as for overall data of Murrah buffaloes. The data pertaining to various economic traits including reproductive and productive characters are utilised for this purpose. The prediction potential of the connectionist models is compared with that of the conventional Multiple Linear Regression (MLR) models. The results revealed that the connectionist models developed in this study seem to be suitable as plausible alternative to conventional MLR models for predicting milk production in Murrah buffaloes. Keywords: Connectionist model, Economic traits, Error back propagation, Generalised regression, Murrah buffalo, Prediction, Radial basis function.
6 Applications of Genetic Algorithms in Agricultural Problems - An Overview
Author: Avnish Kumar Bhatia      Pages: 13-22
Genetic algorithms (GA) are inspired by Darwin?s theory of survival of the fittest in natural genetics. GA is an optimization technique that uses processes of evolutionary biology such as mutation, selection and crossover for artificial evolution towards global optimum in a number of iterations. Various problems in agriculture and livestock management are solved by formulation as optimization problems and hence are candidates for solving with genetic algorithm. Combination of machine learning techniques such as neural networks, fuzzy systems with genetic algorithms has wide applicability in precision farming and green house entailing accuracy of operations. This paper presents a survey of GA applications in solving agricultural problems. Keywords : Genetic algorithm, Agricultural problems.
7 A Reputation Based Service Provider Selection System for Delegation of Job by Farmers
Author: Punam Bedi, Harmeet Kaur and Bhavna Gupta      Pages: 23-32
Selection of a service provider, to perform agriculture jobs, is a major challenge for any farmer. This problem of selection of a service provider to delegate the job is being addressed by a balanced reputation based service provider selection system for farmer. Being agent based system, agents compute the reputation of the service providers present in the e-community based on their past experiences and recommendations collected from their trustworthy acquaintances. Further it is observed that if selection is done only on the basis of the reputation of the service providers then this may introduce delay in the accomplishment of the job because of overloaded reputed service providers. This paper presents a scheme to distribute the work to reputed service providers in such a way that delay in the accomplishment of the job can be minimized. As reputation is a subjective term, so to quantify reputation the concept of Intuitionistic Fuzzy Sets (IFS) is used in this paper. Further the Intuitionistic Fuzzy distances among the recommendations of various trustworthy peers are computed using which the trust on a trustworthy acquaintance is updated. Keywords: Trust, Reputation, Agent, Delegation of job, Intuitionistic Fuzzy Set.
8 Approach for Mining Multiple Patterns from Clusters
Author: Rajni Jain and Alka Arora      Pages: 33-42
Approach for multiple pattern extraction from obtained individual clusters is presented in this paper. Pattern extraction supports the end users in understanding the cluster concept. Pattern discovery approach utilizes the concept of reduct from rough set theory to find out non-significant attributes in a cluster which has no role in pattern formation. These non-significant attributes (reduct) are removed and remaining attributes are ranked for their significance in the cluster. Multiple pattern formulation approach uses ranked attributes to generate concise cluster patterns. Applicability of the approach is demonstrated using soybean disease and zoo datasets from machine learning repository. Objective of applying proposed approach on soybean disease clusters and clusters of zoo animals is to obtain the patterns to describe those clusters. Keywords: Clustering, Data mining, Cluster description, Cluster pattern, Multiple patterns, Reduct, Rough set theory.
9 A Supervised Neural Network Model for Predicting Average Summer Monsoon Rainfall in India
Author: Surajit Chattopadhyay and Goutami Chattopadhyay      Pages: 43-49
Present study aims to develop a predictive model based on artificial neural network (ANN) for the average summer monsoon rainfall amount over India. The dataset made available by the Indian Institute of Tropical Meteorology, Pune, was explored. To develop the predictive model, Backpropagation method with scaled conjugate gradient descent algorithm has been implemented. The ANN model with the said algorithm has been trained thrice to reach a good result. After three runs of the model, it is found that a high prediction yield is available. Finally after rigorous assessment by Willmott?s index, ANN with scaled conjugate gradient descent based Backpropagation algorithm was found to be skillful in predicting average summer monsoon rainfall amount over India. It has been found to be more skillful than non-linear regression in the said prediction task. Keywords: Summer monsoon rainfall, Scaled conjugate descent, Artificial neural network, Prediction, Non-linear regression.
10 Preliminary Study on Prediction of Body Weight from Morphometric Measurements of Goats through ANN Models
Author: A.P. Ruhil, T.V. Raja and R.S. Gandhi      Pages: 51-58
The Artificial Neural Network (ANN) models were developed for prediction of body weight using different linear body measurements in Attappady Black goats of Kerala, India. Data on body weight and body measurements recorded on 919 female goats from its breeding tract were used for the study. The whole data was classified into four age groups viz., 0-3, 3-6, 6-12, and above 12 months. From the whole data sets of different age groups 75 per cent were used to train the neural network model and remaining 25 per cent were used to test the model. Three different morphometric measurements viz., chest girth, body length and height at withers were used as input variables and body weight was considered as output variable. The network architecture used was a multilayer feed forward network with back propagation of error learning mechanism. The accuracy of prediction of body weight from ANNs analysis was higher when compared to MRA indicating that the ANN models were able to describe more variation in live weight. Maximum prediction accuracy (77.19%) and minimum SD ratio (0.4838) was noticed for 0-3 months age groups and the RMSE was maximum for >12 months age groups (2.7255). The phenotypic correlations between actual and predicted body weights were positive, and highly significant (P < 0.01) at all the age groups. The predicted body weights were perfectly acceptable when compared to the actual body weights. It was concluded that artificial neural network (ANN) models could be used as an alternative to traditional MRA for estimating the live weight using linear body measurements. Keywords: Artificial neural networks, Attappady Black goat, Body measurements, Body weight prediction, Multiple regression analysis.
11 Attribution Analysis and Classification of miRNAs
Author: A.K. Mishra and D.K. Lobiyal      Pages: 59-70
Several models for miRNA prediction based on attributes of known miRNAs have been developed. These models are based on different sets of attributes. However, only limited attempts have been made in exploring dominating attributes to reduce the complexity of the model and increase the classification accuracy. To the best of our knowledge statistical techniques for attribute selection have not been applied for miRNA attribute analysis. miRNA prediction, generally consider attributes from few model organisms widely used by researchers from biological sciences. Further, most of the models focus more on training algorithms to improve the prediction accuracy rather than attributes relevance analysis. In this paper we have derived 14 attributes for precursors, 9 attributes for mature miRNAs and 20 attributes in combination from both precursors and mature miRNA for relevance analysis and classification of four hexapode species-Apis mellifera, Bombyx moori, Anopheles gambiae and Drosophila Melanogaster. Dominating feature extraction was done using different machine learning techniques from a set of known miRNA sequences for the above mentioned species using PCA, Infogain, SVM attribute analysis, Cfs subset, Consistency subset evaluation and Chi squared analysis. The results are encouraging since the essential attributes selected here are biologically significant. These attributes can be used in deriving rules for miRNA identification. The performance measures on training and test datasets are quite satisfactory. The results obtained from experiments clearly show that our model gives high precision and recall for all the four species in all combinations. Future developments ought to focus on the need to establish more accurate models using sophisticated algorithms of artificial intelligence techniques and rule based mining approach. Keywords: miRNA, Attribute, Classification, Relevance, Prediction.
12 Establishment of Castor Core Collection Utilizing Self-Organizing Mapping (SOM) Networks
Author: C. Sarada and K. Anjani      Pages: 71-78
A core collection can be defined as a representative sample of entire germplasm collection with minimum repetitiveness and maximum genetic diversity of a crop species and its relatives. The success of development of a most representative core collection mainly depends on non-overlapping grouping of whole collection. In the present study, a promising method viz., Self Organizing Mapping (SOM) network clustering technique was applied, which was first time attempted in establishment of core collection in a crop species. An attempt was made to compare SOM with clustering methods viz., Ward?s and K-means clustering to understand the superiority of SOM over these two methods in forming castor core representative of whole collection. Forty experimental cores were constructed using these clustering methods as well two clustering algorithms ( single and two stage) and two allocation methods, viz., proportional and logarithmic methods. Three sample sizes representing 10 per cent, 15 per cent and 20 per cent of total collection were drawn, and a fourth sample size of 524 accession based on progresss was made. Thus formed experimental cores were evaluated based on the four parameters viz., mean difference percentage (MD), variance difference percentage (VD), coincidence rate percentage (CR) and variable rate percentage (VR). The results indicated that SOM method performed better as compared to Ward?s and K-means clustering methods conserving maximum diversity existing in the whole germplasm collection.
13 Expert System for Disease Diagnosis in Soybean-ESDDS
Author: Savita Kolhe, Raj Kamal, Harvinder S. Saini and G.K. Gupta      Pages: 79-88
The paper describes the development of expert system for disease diagnosis in soybean. It explains the methodology for development of knowledge-based expert system. The different components of the expert system are explained. The inference engine of the system has been developed as Object-Oriented (O-O) inference model using O-O programming. The O-O inference model involves disease management case studies, fuzzy logic algebra, rule-promotion strategy, rule-patterns experience and statistical methods in the form of objects as an analytical tool and predicts accurate disease. The knowledge base is implemented in the form of relational database using SQL Server. The user interface of the software is modeled based on 3-tier architecture design using ASP.NET. The system evaluation study includes verification and validation processes. The interface of system is evaluated on ten design features. The validation process was conducted by a team of twenty agriculture researchers with different level of experiences in agriculture. The evaluators were highly satisfied with the user interface and found it user-friendly. The overall average rating given by the users was more than 8 on a 1-10 point scale. The inference drawing process of inference engine was significantly improved by application of new fuzzy-logic rule-promotion approach. The number of successful diagnoses was increased from 65% upto 92.4% by applying new approach. The system as a whole is a powerful means for transfer of soybean pathological technologies to practices over the web. The inference model developed in this work can be reused for other crops also. Keywords: Disease diagnosis, Expert system, Fuzzy-logic, Inference engine, Knowledge base, Soybean disease diagnosis.
14 Extracting Concepts using Linguistic Ontology in Agriculture Domain
Author: Aditi Sharma, Nidhi Malik and Vajenti Mala      Pages: 89-96
With the widespread and increasing availability of text documents in electronic form and need for managing the information that resides in the vast amount of available text documents, the field of Text Mining is receiving a lot of attention. For the same reason text mining is also becoming important task in agriculture domain. Traditional text mining was based on keyword based extraction which has some limitations , as this method does not consider inherent semantic relationship among the words. Semantic text mining can overcome these limitations. Semantic text mining aims at discovering the hidden information from the text documents based on relationships of the terms occurring in them. Linguistic Ontology is one of the most widely used tools for semantic text mining. The objective of this paper is to highlight the role of linguistic ontology in mining textual information with a focus on agriculture domain. Specifically, we present an algorithm for extracting concept clusters from text documents using WordNet ontology. We have taken documents from the agriculture field and performed experiments on them. The results are encouraging and encourage us to explore this area further. Keywords: Information retrieval, Semantic relations, Concept cluster, WordNet ontology, Concept based information retrieval, Agriculture.
15 An Intelligent Semantic Search Engine with Cross Linguistic Support
Author: S.D. Samantaray      Pages: 97-106
In this paper, an intelligent semantic based search engine has been presented that can be used as a multilingual platform for different search queries. It retrieves those results pages also which don?t have directly the keywords but contains the synonyms or related words. In response to a query for the word ?soil? it will also retrieve web pages which don?t have directly the word ?soil? but have the semantically related words such as ?land?, ?ground?, ?earth?, ?loam? etc. As a cross lingual support it retrieves the pages of other languages which have semantic presence of the searched keyword. These pages are presented to user in the original query language after necessary translation/transliteration. Spell-check support is also provided for giving suggestions in respect of misspelled queries. For the search engine implementation, JSP (Java Server Pages) has been used to integrate the Front-end (webpage) and Back-end (crawler and indexer). LUCENE (a very popular open source information retrieval library) and Word Net library has been used for getting the synonyms list of the queried words. Keywords: Intelligent search, Natural language processing,Semantic web, Artificial intelligence, Intelligent systems, Cross linguality.