Artificial Neural Networks is a popular choice in optimization tasks for a number of applications such as approximations, regression, classification. The speed of training of Artificial Neural Networks is sensitive to weight initialization. A new interval-based weight initialization method I-WT is proposed to improve the convergence rate in artificial neural networks. The lower and upper bound is used, so as to model the uncertainty (such as noise in case of measured data), in the estimation of free parameters, i.e., weights. The novelty of this approach lies in the exploitation of dependency of weight update on the derivative of activation function to determine the optimal interval for weight initialization, that reduce the chance of saturation to only 6

During the early stages of the life cycle development process for software, the developer mainly makes use of the fault prediction process for the development of different modules. These modules help in detecting faulty modules and classes. Further, this process also helps in determining the modules which require a high level of refactoring during the maintenance stage. The objective of this research is to classify faults and to explore the usability of Factor Analysis with Regression (FAWR) which drastically ameliorate the system performance. A review of recent studies performed that uses the different fault prediction techniques. To direct this research, two research questions (RQ) are defined, one related to the integration of techniques to enhance the development of fault prediction model, and another is to check the technique to overcome the limitations of old methods. To answer these RQs, FAWR techniques are used for predicting faults. To assess the quality of the technique, two experiments were conducted. Results show that FAWR is the better performing method among the two prediction methods investigated. The results proved that the prediction capability of FAWR technique is significantly better. Factorization method is able to classify a module whether it is fault-prone or not. The constructed models use to estimate the proneness of faults surpass the standard regression models. The system evaluations indicate that the reduction of terms results in the betterment of outcomes. Moreover, the consideration of FAWR is a significant technique for the prediction of faults in software.

Activation functions play a major role in determining the learning speed and generalization capability of feed-forward artificial neural networks. In this paper, an empirical comparison of eight activation functions is reported on 12 function approximation problems. The study allows us to assert that the sigmoidal class of activation functions performed much better than the non-sigmoidal class of activation functions. Out of the six non-sigmoidal activation function, one activation function called the sigmoidal-weighted linear unit is identified as outperforming all other non-sigmoidal activation functions.

In this paper, a bounded, continuous, differentiable and non-monotonically increasing function is proposed (σ05). The efficiency and efficacy of using this function as an activation function for hidden layer nodes of FFANNs is demonstrated on five function approximation problems. The proposed function is compared against four bounded, continuous, differentiable, and non-monotonically increasing function (also called sigmoid functions). The results demonstrate that the usage of the proposed activation function as an activation function leads to creation of networks that generalize better than FFANNs using other (sigmoidal) activation function(s).

Feedforward neural networks are known to exhibit universal approximation property, i.e., these networks can effectively approximate any function arbitrary well, provided there is one hidden layer with sufficient number of hidden nodes, and some non-linearity at the hidden layer nodes. The property of universal approximation in these networks contributed to wide range of application, remarkably in the area of predictive modelling. However, these networks have a major drawback of slow convergence. In this work, we proposed a new weight initialization algorithm to improve the convergence speed in feedforward networks using statistical analysis of system, based on the set of predefined assumptions. The proposed algorithm is compared to other established weight initialization algorithm for predictive models, and is expected to perform better than other algorithms.

The most commonly used activation functions in the field of artificial neural networks is the class of sigmoidal activation functions which are bounded, continuous, differentiable and monotonically increasing function. The universal approximation results for feed-forward artificial neural networks allow activation functions to be any arbitrary non-polynomial function. In this paper the properties of a bounded, continuous, differentiable and non-monotone function is described. The efficacy and efficiency of using this non-monotone function as activation function is demonstrated on five benchmark learning task. The non-monotone activation function is compared with four generally used activation functions on the benchmark tasks. Results demonstrate that the networks using the non-monotone activation function at hidden layer nodes out-perform the other 4 sigmoidal activation function using networks.

Recently, Internet of Things (IoT) and multimedia are gaining popularity because of their usages in various applications. Numerous sensors and automated devices are generating huge volumes of data. Therefore, it is required to efficiently and effectively analyze this voluminous data. It can be achieved by using appropriate machine leaning techniques such as clustering. Among the clustering techniques, the k-means method/algorithm is one of the simplest, effective and commonly used methods. For making the cluster, it uses a measure of similarity/distance among the data observations. Nearby/similar data observations are placed within the same cluster whereas distant/dis-similar data observations are placed in other clusters. Hence, the similarity/distance metric plays a major role on the performance and accuracy of the k-means. Therefore, using an appropriate similarity/distance metric, the performance and accuracy of the k-means can be improved. K-means algorithm is majorly implemented using Euclidean distance metric. With the objective to explore the better and/or alternate similarity/distance metric(s) for k-means, a case study, based on empirical evaluation, of thirteen different similarity/distance metrics on six well-known datasets is performed and presented in this paper. By using the efficient and effective similarity / distance metrics, the performance and accuracy of the k-means algorithm can be improved which leads to formation of good clusters of various data observations or things or images etc. The results of the empirical study are analyzed and compared on the basis of widely used statistical clustering evaluation/validation measures. Based on the comparative results, these metrics are assigned with the ranks. Overall, the results demonstrate that Manhattan and Minkowski distance metrics gives better results for k-means algorithm.

In this paper a new weight initialization scheme is proposed. The proposed scheme, distributes the biases / thresholds at a equal intervals in an interval ( - λ, λ); while the weights are distributed uniformly in the same interval. The value of the λ is chosen such that the expected value of the net inputs is 0 while the variance is 1. On a set of 10 tasks (5 function approximation and 5 real life benchmark regression tasks), the proposed weight initialization routine was compared to three existing weight initialization routine. From the results obtained we may infer that the proposed weight initialization is almost always better than these three existing routines and is never worse on the basis of generalization performance (result over the test data set).

SEP is one of the most simple and popular protocol for clustered, heterogeneous wireless sensor networks (WSNs). Since its inception, it has either been a direct or indirect source of reference for development of new protocols for heterogeneous WSNs. There are certain approximations and assumptions in SEP. In this paper we have explored the consequences of not considering some of these approximations. We have re-evaluated expression for optimum number of clusters, identified factors determining value of optimum number of clusters and analysed their effect by varying them. Analytically we have found that below a threshold distance between node(s) and base station, clustering is not required, or there can be a mixed approach to it. Simulation results show substantial improvement in lifetime and throughput.

For finding the performance of neural network choice of training algorithm is very important. Various algorithms are available and are used for testing the performance. In this paper we have used twelve variants of Back propagation algorithm for training & testing. We have used four different weight initialization methods namely random, Nguyen-Widrow and two proposed weight initialization methods. We have taken ten different data sets. We have compared the results obtained by applying all the twelve algorithms and found that our proposed methods are giving better results when compared with random and Nguyen-Widrow weight initialization methods. We have found that we obtained best results when air quality data set was taken as an input and scaled conjugate gradient back propagation algorithm as training algorithm out of all the twelve training algorithms taken during experimentation.

k-means is an effective and efficient clustering algorithm. It uses distance/similarity metric to find out the distance/similarity among the data objects. The objects which are closer/similar to each other are assigned to the same cluster where as distant/dissimilar objects are assigned to different clusters. Most of the implementations of k-means are based on Euclidean/Squared Euclidean distance metrics. In order to find out the possibility of different distance/similarity metrics to be used with k-means algorithm, an empirical evaluation has been performed. In this paper, accuracy, performance and reliability of 13 different distance/similarity measures over 6 different variations of data using k-means algorithm have been compared based on empirical evaluation on well-known benchmark IRIS data set. Accuracy is measured in terms of similarity of cluster assignment between ground truth and machine clustering. Performance is measured in terms of the number of iterations used for convergence of the final cluster assignment. Reliability is measured on the basis of correctness of the cluster assignment.

Data mining plays an important role in various human activities because it extracts the unknown useful patterns (or knowledge). Due to its capabilities, data mining become an essential task in large number of application domains such as banking, retail, medical, insurance, bioinformatics, etc. To take a holistic view of the research trends in the area of data mining, a comprehensive survey is presented in this paper. This paper presents a systematic and comprehensive survey of various data mining tasks and techniques. Further, various real-life applications of data mining are presented in this paper. The challenges and issues in area of data mining research are also presented in this paper.

Choice of initial centroids has a major impact on the performance and accuracy of k-means algorithm to group the data objects into various clusters. In basic k-means, pure arbitrary choice of initial centroids lead to construction of different clusters in every run and consequently affects the performance and accuracy of it. To date, several attempts have been made by the researchers to increase the performance and accuracy of it. However, scope of improvement still exists in this area. Therefore, a new approach to initialize centroids for k-means is proposed in this paper on the basis of the concept to choose the well separated data-objects as initial cluster centroids instead of pure arbitrary selection. As a consequence, it leads to higher probability of closeness of the chosen centroids to the final cluster centroids. The proposed algorithm is empirically assessed on 6 different well-known datasets. The results confirms that the proposed approach is considerably better than the pure arbitrary selection of centroids.

Weight initialization of sigmoidal feed forward artificial neural network (SFFANN) and the Convolutional neural networks (CNN) has been a known factor which affects the learning abilities of the neural network. The uniform random weight initialization approach has been quite often used as the conventional network weight initial technique, due to its simplicity. However, various researches have shown that the random technique may not be the ideal choice of weight initialization for these neural networks. In this work, we analyze two separate chaotic functions and explore the possibilities of these being used as the weight initialization methods against the conventional random initialization technique for SFANNs as well as for the CNNs. For the SFFANNs, this analysis were done over 8 function approximation problems chosen for experimentation. The mean test error values along with a two sample t-test results strongly suggest that the Chebyshev chaotic map based weight initialization technique outperforms the conventional random initialization technique for most of the problems under consideration and hence may be used as an alternative weight initialization technique for the SFFANNs. For the CNN experiment, the MNIST dataset was used for analyzing the performance of the random and the Chebyshev based initialization scheme. Results strongly support the use of the Chebyshev chaotic map based initialization scheme as an alternate to the conventional random initialization.

The world of software systems is amplified with the changing environment magnifying the demand for quality software. Software fault prediction is a requisite activity ensuring the development of economic, efficient and quality software. It is the procedure for the development of models which help to identify faults in modules during early phases of software development lifecycle. Software fault prediction is one of the most prevalent research disciplines. The existing study in this domain includes numerous modelling techniques and software metrics for the early predictions of software faults. This paper aims to explore some of the prominent studies for software fault prediction in the existing literature. In this paper, software fault prediction papers since 1990 to 2017 are investigated. The paper includes the analysis of the studies having empirical validation and a good source of publication. The paper reflects the methods, metrics, and datasets available in the literature for software fault prediction. In addition, the modelling techniques based on traditional and computational intelligence-based methods are also reviewed. This paper is an endeavour to assemble the existing techniques and metrics of software fault prediction with a motive to assist researchers for easy evaluation of suitable metrics for their own research scenarios.

In software fault prediction, before beginning with the real software testing process, fault-prone software modules are identified with the help of various properties related to the software project. This leads to the achievement of minimal cost apart from the desired software quality. In this study, Object oriented metrics were used to find the main factors. The technique used to find the important predictors is Factor analysis (FA) and regression is used to analyze the goodness of fit of different models drawn from previous studies. Following this, identification of various factors associated with the software fault prediction process is done. In addition, a robust model is developed on the basis of the distinct identified factors. The research work is the extension of our previous work. This research provides a novel approach of different aspects of the software fault prediction process and also assists in discovering the different types of problems associated with the term i.e. software fault prediction. Towards the end, the statistical comparison and the challenges of the paper are discussed along with the future guidelines of the study.

Training in feedforward networks is adaptation of weights of network which makes weight initialization an important factor in determining the training speed in network. This work aims at proposing a new weight initialization technique, wherein the weights are initialized in the useful range in threshold function. It further initializes weights to each hidden node in a region which alternately contracts and expands statistically. The proposed weight initialization is compared to the conventional random weight initialization. The proposed weight initialization is expected to perform better than the random weight initialization.

The selection of the initial network weights has been a known key aspect affecting the convergence of sigmoidal activation function-based artificial neural networks. In this paper, a new network initialization scheme has been proposed that initializes the network weights such that activation functions in the network are not saturated initially. The proposed method ensures that the initial outputs of the hidden neurons are in the active region which positively impacts the network’s rate of convergence. Unlike most of the earlier proposed initialization schemes, this method does not depend on architectural parameters like the size of the input layer or the hidden layer. The performance of the proposed scheme has been compared against eight well-known weight initialization routines over six benchmark real-world problems. Results show that the proposed weight initialization routine enables the network to achieve better performance within the same count of network training epochs. A right-tailed t-test also shows that our proposed scheme is significantly better in most of the cases against the other techniques or statistically similar in a few cases but never underperforms. Hence, it may be considered as a strong alternative to the conventional neural network initialization techniques.

The success of the Sigmoidal Feedforward Networks in the solution of complex learning task can be attributed to their Universal Approximation Property. These networks are trained using non-linear iterative optimization method (of first-order or second-order) to solve a learning task. The convergence rate in Sigmoidal Feedforward Network training is affected by the initial choice of weights, therefore, in this paper, we propose two new weight initialization routines (Routine-1 and Routine-2) using characteristics of input and output data and property of activation function. Routine-1 uses the linear dependency of weight update step size on derivative of activation function and thus, initialize weights and bias to activate the activation function region near zero (input), where the derivative is maximum, therefore, increasing the weight update step size, and hence, the convergence speed. The same principle is used to derive Routine-2, that initialize weights and bias to activate distinct point in the significant range of activation function (where significant range defines the non-saturated region in activation function), such that, each node evolves independently of each other, and act as distinct feature identifier. Initializing weights in significant range reduces chances of (hidden) nodes getting stuck in saturated state. The networks initialized using proposed routines has higher convergence and higher probability to achieve deeper minima. The efficiency of proposed routines is evaluated by comparing them to conventional random weight initialization routine and 11 weight initialization routines proposed in literature (4 well established routines and 7 recently proposed routines) for several benchmark problems. The proposed routine is also tested for larger networks sizes and larger datasets such as MNIST. The results show that the performance of proposed routines is better than conventional random weight initialization routine and 11 established weight initialization routines.

Grouping of sensor nodes in clusters has several advantages including energy efficiency, network scalability, and efficient data aggregation. Many clustering protocols have been developed till date promising better energy efficiency in comparison with others. In this paper, we have surveyed important clustering techniques with a focus on the estimation of optimum number of clusters. We have also presented a case study on LEACH protocol, suggesting that under certain conditions clustering is not a wise solution, a non-clustered network or a network with mixed approach can give better result. Experimental results show significant improvement in lifetime and throughput.