Amit Ganatra

Searching Most Efficient Neural Network Architecture Using Akaike's Information Criterion (AIC)

The problem of model selection is considerably important for acquiring higher levels of generalization capability in supervised learning. Neural networks are commonly used networks in many engineering applications due to its better generalization property. An ensemble neural network algorithm is proposed based on the Akaike information criterion (AIC). Ecologists have long relied on hypothesis testing to include or exclude variables in models, although the conclusions often depend on the approach used. The advent of methods based on information theory, also known as information-theoretic approaches, has changed the way we look at model selection The Akaike information criterion (AIC) has been successfully used in model selection. It is not easy to decide the optimal size of the neural network because of its strong nonlinearity. We discuss problems with well used information and propose a model selection method.

Forecasting Employee Retention Probability Using Back Propagation Neural Network Algorithm

The Artificial neural networks are relatively crude electronic networks of neurons based on the neural structure of the brain. It process the records one at a time, and learn by comparing their prediction of the record with the known actual record. The errors from the initial prediction of the first record is fed back into the network, and used to modify the networks algorithm the second time around and so on for many iterations. The goal is to identify potential employees who are likely to stay with the organization during the next year based on previous year data. Neural networks can help organizations to properly address the issue. To solve this problem a neural network should be trained to perform correct classification between employees. After the network has been properly trained, it can be used to identify employees who intent to leave and take the appropriate measures to retain them

A Unique Word Prediction System for Text Entry in Hindi

Word prediction is very effective technique for improving efficiency of entering text. Current word prediction systems predict a word if and only if a user has not made mistake in the starting of some characters of the word. This is more applicable for Indian languages, which have a large set of characters, alphabets, words with complex characters and inflections, phonetically similar sets of characters, etc. Therefore, there is a requirement for development of better word prediction. For existing systems, till now N-Gram approach is used. N-Gram approach considers only sequence of words in given sentence. It doesnt consider structure and grammar of the sentence. New approach is to use Syntactic N-Gram approach. Sn-Grams are differing from traditional n-grams in the way of which elements are considered as the neighbors. Sn-Grams consider Grammar in making prediction. So they are less arbitrary in making predictions.

Efficient constraint-based Sequential Pattern Mining (SPM) algorithm to understand customers’ buying behaviour from time stamp-based sequence dataset

Abstract Business Strategies are formulated based on an understanding of customer needs. This requires development of a strategy to understand customer behaviour and buying patterns, both current and future. This involves understanding, first how an organization currently understands customer needs and second predicting future trends to drive growth. This article focuses on purchase trend of customer, where timing of purchase is more important than association of item to be purchased, and which can be found out with Sequential Pattern Mining (SPM) methods. Conventional SPM algorithms worked purely on frequency identifying patterns that were more frequent but suffering from challenges like generation of huge number of uninteresting patterns, lack of user’s interested patterns, rare item problem, etc. Article attempts a solution through development of a SPM algorithm based on various constraints like Gap, Compactness, Item, Recency, Profitability and Length along with Frequency constraint. Incorporation of six additional constraints is as well to ensure that all patterns are recently active (Recency), active for certain time span (Compactness), profitable and indicative of next timeline for purchase (Length―Item―Gap). The article also attempts to throw light on how proposed Constraint-based Prefix Span algorithm is helpful to understand buying behaviour of customer which is in formative stage.

Performance evaluation PL330 DMA controller for bulk data transfer in Zynq SoC

This paper provides performance evaluation of PL330 DMA in Zynq SoC based device. Direct Memory Access is the feature that allows computer hardware to access system memory for data movement in bulk without CPU intervention. The I/O devices operate at a slower speed than CPU, but using DMA the CPU can be available for performing other computing tasks while data is transferred, as CPU has to only initiate the read/write of data. The direction of transfer can be from device-to-memory, memory-to-device, memory-to-memory and device-to-device. This paper describes the hardware setup and sequence of operations for transfer of bulk volume data at high speed using PL330 DMA controller in Zynq SoC based system.

Techniques and Challenges in Building Intelligent Systems: Anomaly Detection in Camera Surveillance

Security is tedious, complex and tough job in today’s digitized world. An attempt is made to study and propose an intelligent system for surveillance. Surveillance camera systems are used for monitoring and controlling the security. Anomaly detection techniques are proposed for designing the intelligent control system. In the paper challenges in detection and processing of anomaly in surveillance systems are discussed and analyzed. Major components related to an anomaly detection technique of camera control system are proposed in the paper. Surveillance data is generated through camera, and then this data is transmitted over the network to the storage. Processing is to be done on real time basis and if there is any anomaly detected, the system must produce an alert. This paper is an attempt to study soft computing approaches for anomaly detection.

Information Fusion (Ensemble System) in Data Warehousing & Data Mining

Machine based systems cant keep up with the task of organizing the data in an up-to-date manner unless and until the data acquired is being planned or scheduled and managed in an appropriate manner. Todays datasets start as small chunk of information and grow exponentially over a period of time. Once the size is extremely large it becomes difficult to make decisions and to predict consistently and correctly from the datasets. Most of predictions do not hold true, if proper balancing and diversification in terms of certain conditions and parameters is not done. The present state has focused public attention in terms of making and combining the predictions from the available data i.e. analyzing the current (and past) data to make predictions with increasing predictive accuracy of the overall system. So, keeping these considerations in mind there is a need for the better concept (component) for Information Fusion to combine it with a solid theory in support and foundation. AdaBoost could be very useful with feature selection, especially when considering that it has solid theoretical foundation. Here, Genetic Algorithms are being used to select relevant features from large datasets along with Evaluation techniques. This can further be enhanced by using multiclassifier approach. The central objective is to develop the system that provides approximately 3-5% performance improvement at least over similar existing techniques.

Incorporating boundary value concept and recency constraint to capture emerging trends in time stamp based sequence dataset

Sequential Pattern Mining (SPM) is an important subject to focuses on the current purchase scenario. Current research of SPM mostly emphasis on items or itemset which are frequently purchase in certain order but not focused on the purchase which will be potentially strong for future. In this paper, we have tried to focus on Emerging Patterns (EPs) based on significant change of its support values. Proposed algorithm worked on significant change of boundary value support threshold. There are some patterns which lies on boundary, due to its little law amount of frequency they are rejected, those patterns are not in consideration presently but which having similar kind of nature as discovered patterns. Incorporation of manipulation in Boundary value and Recency constraint in traditional SPM algorithm - PrefixSpan, helps to discover EPs and distinguish between actual infrequent patterns as well patterns suffering from little low support threshold from immense database. Proposed algorithm effectively discover the current spending patterns of customers and recent trends, which will allow decision maker to detect in a large database potential changes of customer preference, and provide as early as possible products and services desired by the customers to expand and retain business.

The Shibir: Iteaching in villages — A CHARUSAT initiative for IT penetration in rural areas

Information Technology is becoming fundamental to the operation of business and government. Growth in jobs related to IT is expected to outpace traditional industries. In a globalizing economy, Computer and Information Technology are being perceived as a new engine of economic and industrial growth. Cities and Metros are growing at an unprecedented pace with the use of IT. India is a country of villages where the development has not reached till date. Rural development implies both the economic betterment of people as well as greater social transformation. This paper presents a live case study of our own university which has been involved in teaching IT to villagers for betterment of their lives.

Performance analysis of classification techniques using different parameters

Data mining technique is the process of analyzing data from different perspectives and summarizing it into useful information. Classification refers to the data mining problem of attempting to predict the category of categorical data by building a model based on some predictor variables. The goal of data classification is to organize and categorize data in distinct classes. It does not require any priori knowledge of the class statistical distribution in data sources. ANN can be trained to distinguish the criteria used to classify and it can do so in a generalized manner allowing successful classification of new inputs not used during training. Back propagation as a training algorithm for ANN works well for classification. This Paper shows the issue of improving the fitness of BPN algorithm and the performance analysis of various classification techniques like Naive Bayes, Bayesian network, Support Vector Machine and GABPN discussed.