Amit Thakkar

Boost a Weak Learner to a Strong Learner Using Ensemble System Approach

The goal of classification learning is to develop a model that separates the data into the different classes, with the aim of classifying new examples in the future. A weak learner is one which takes labeled training examples and produces a classifier which can label test examples more accurately than random guessing. When such weak learner is used directly for classification task then it may not give the better prediction accuracy, due to the limitation and simplicity of single classifier system. On the other hand, multiple classifier systems often known as ensemble based systems, have shown to produce favorable results compared to single-classifier systems. Boosting is one of the most important recent developments in ensemble system, which works by sequentially applying a classification algorithm to re-weighted versions of the training data and then taking a weighted majority vote of the sequence of classifiers. Our experiments demonstrate the underlying weak learners ability to achieve a fairly low error rate on the testing data, as well as the boosting algorithms ability to reduce the error rate of the weak learner. In our experiment we have used decision stump as a weak learner (classifier) and using the boosting approach, the result demonstrates the improvement in the classifiers accuracy.

Ranking of Classifiers based on Dataset Characteristics using Active Meta Learning

Classification is a machine learning technique which is used to categorize the different input patterns into different classes. To select the best classifier for a given dataset is one of the critical issues in Classification. Using cross-validation approach, it is possible to apply candidate algorithms on a given dataset and best classifier is selected by considering various evaluation measures of Classification. But computational cost is significant. Meta Learning automates this process by acquiring knowledge in form of Meta-features and performance information of candidate algorithm on datasets and creates a Meta Knowledge Base. Once Meta Knowledge Base is generated, system uses k-Nearest Neighbor as a Meta Learner that identifies the most similar datasets to new dataset. But generation of Meta Example is a costly process due to a large number of candidate algorithms and datasets with different characteristics involved. So Active Learning is incorporated into Meta Learning System that reduces generation of Meta example and at the same time maintaining performance of candidate algorithms. Once the training phase is completed based on Active Meta Learning approach, ranking is provided based on Success Rate Ratio (SRR) method that considers accuracy as a performance evaluation measure.

A State of Art Survey on Shilling Attack in Collaborative Filtering Based Recommendation System

Recommendation system is a special type of information filtering system that attempts to present information/objects that are likely to the interest of user. Any organization, provides correct recommendation is necessary for maintain the trust of their customers. Collaborative filtering based algorithms are most widely used algorithms for recommendation system. However, recommender systems supported collaborative filtering are known to be extremely prone to attacks. Attackers will insert biased profile information or fake profile to have a big impact on the recommendations made. This paper provide survey on effect of shilling attack in recommendation systems, types of attack, knowledge required and existing shilling attack detection methods.

Correlation Based Anonymization Using Generalization and Suppression for Disclosure Problems

Huge volume of detailed personal data is regularly collected and sharing of these data is proved to be beneficial for data mining application. Data that include shopping habits, criminal records, credit records and medical history are very necessary for an organization to perform analysis and predict the trends and patterns, but it may prevent the data owners from sharing the data because of many privacy regulations. In order to share data while preserving privacy, data owner must come up with a solution which achieves the dual goal of privacy preservation as well as accurate data mining result. In this paper k-Anonymity based approach is used to provide privacy to individual data by masking the attribute values using generalization and suppression. Due to some drawbacks of the existing model, it needs to be modified to fulfill the goal. Proposed model tries to prevent data disclosure problem by using correlation coefficient which estimates amount of correlation between attributes and helps to automate the attribute selection process for generalization and suppression. The main aim of proposed model is to increase the Privacy Gain and to maintain the accuracy of the data after anonymization.

Publish/Subscribe Mechanism for IoT: A Survey of Event Matching Algorithms and Open Research Challenges

The number of sensors getting deployed around the world is increasing due to emergence of Internet of Things. It provides advanced connectivity and communication between devices which goes beyond machine-to-machine communication. Huge amount of data is expected to be generated from different locations that will be aggregated, processed and forwarded very quickly. Publish/Subscribe mechanism is powerful way to allow IoT devices to connect and communicate with each other. One of the major bottlenecks in using Publish/Subscribe systems is the efficiency of filtering incoming message. This is a very challenging problem because in a Publish/Subscribe system the number of subscriptions can be very large. There are quite a few event matching algorithms proposed in the literature to improve its efficiency. The aim of this research paper is to study and analyze how existing approaches ensure fundamental event matching requirements and discuss the open challenges and future work in the area.

Link-based classification for Multi-Relational database

Classification is most popular data mining tasks with a wide range of applications. As converting data from multiple relations into single flat relation usually causes many problems so classification task across multiple database relations becomes challenging task. It is counterproductive to convert multi-relational data into single flat table because such conversion may lead to the generation of huge relation and lose of essential semantic information. In this paper we propose two algorithms for Multi-Relational Classification (MRC). To take advantage of linkage relationship and to link target table with different tables, a semantic relationship graph (SRG) is used. In First approach we have used Naïve Bayesian Combination to combine heterogeneous classifiers result to get class label. This will classify the instance accurately and efficiently. Second approach is Multi-Relational Classification using Decision Template (DT). Decision profile is created to combine heterogeneous classifiers output. Based on similarity measure decision template and decision profile is compared to get final output. DT takes contribution of each classifiers output rather than class-conscious. So classification accuracy is improved.

Neural network with deep learning architectures

Abstract Deep Learning is a field included in to Artificial Intelligence. It allows computational models to learn multiple levels of abstraction with multiple processing layers. This Artificial Neural Networks gives state-of-art performance in various fields like Computer Vision, Speech recognition and different domain like bioinformatics. There are mainly three architectures of Deep Learning Convolution Neural Network, Deep Neural Network and Recurrent Neural Network which provides the higher level of representation of data at each next layer. Deep Learning is required to classify high dimensional data like images, audio, video and biological data.

Comprehensive and Evolution Study Focusing on Comparative Analysis of Automatic Text Summarization

In the escalating trend of atomization and online information, text summarization bolster in perceiving textual information in the form of summary. It’s highly tedious for human beings to manually summarize large documents of text. In this paper, a study on abstractive and extractive content rundown strategies has been displayed. In Extractive Text Summarization it talk about TF-IDF, Cluster based, Graph theory, Machine learning, Latent Semantic Analysis (LSA) and Fuzzy logic approaches. Abstractive rundown techniques are ordered into two classes i.e. Structured based approach and Semantic based approach. In Structure Based approach it talk about Tree based, Template based, Ontology based, Lead & Phase based and Rule based method. In Semantic Based Approach it talks about Multimodal semantic, Informative item based and Semantic graph based method. The central idea of this method has been elaborated further, apart from idea, the advantages and disadvantages of these methods have been procured.

Recommendation System for Improvement in Post Harvesting of Horticulture Crops

Horticulture includes tropical and subtropical fruits, vegetables, spices, flowers, medicinal and aromatic plants. Horticulture sector is a major growth of Indian Agriculture. India is second largest producer of fruits and vegetables in the world. But the post-harvest loss is because of weak supply chain entities like storage facilities, bad transportation facility, market facility, and not proper packaging, not use of modern techniques, not proper post-harvest management. Due to post harvest loss actual need of fruits does not satisfy and so that need to import the fruit from outside the country. If import of fruit is higher than the export then it will impact on balance of payment of India, value goes negative. Post-harvest loss indirectly affect on our Indian Economy. By using modern technologies post-harvest loss can be reduced. For example, Geographic Information System (GIS) can be used for analysis of spatial data and helps also in decision making in problem. Location based recommendation system will also help to recommend the location of cold storages and establishment of new cold storages.

Improving efficiency of heterogeneous multi relational classification by choosing efficient classifiers using ratio of success rate and time

AbstractTraditional data mining algorithms will not work efficiently for most of the real world applications where the data is stored in relational format. Useful patterns can certainly be extracted from multiple relations using an existing traditional learning algorithm of data mining, but it would involve a lot of complexity. So there is a need of a multi relational classification, which analyzes relational data and predicts unknown patterns automatically. Moreover the performances of existing relational classifiers are limited, because the existing algorithms are not able to use different classifiers based on characteristics of different relations. The goal of the proposed approach is to select appropriate classifiers based on characteristics of different relations in the relational database to improve the overall performance without affecting the running time. So multi criteria classifier selection function based on ratio of accuracy and running time is used to select the most efficient classifier usi...