Abstract
A number of ANN methods are used, but BP is the most commonly used algorithms to train ANNs by using the gradient descent method. Two main problems which exist in BP are slow convergence and local minima. To overcome these existing problems, global search techniques are used. This research work proposed new hybrid flower pollination based back propagation HFPBP with a modified activation function and FPBP algorithm with log-sigmoid activation function. The proposed HFPBP and FPBP algorithm search within the search space first and finds the best sub-search space. The exploration method followed in the proposed HFPBP and FPBP allows it to converge to a global optimum solution with more efficiency than the standard BPNN. The results obtained from proposed algorithms are evaluated and compared on three benchmark classification datasets, Thyroid, diabetes, and glass with standard BPNN, ABCNN, and ABC-BP algorithms. The simulation results obtained from the algorithms show that the proposed algorithm performance is better in terms of lowest MSE (0.0005) and high accuracy (99.97%).
Top1. Introduction
A back-propagation (BP) neural network can resolve complex arbitrary nonlinear planning problems; therefore, it can be applied to a varied problem. However, as the model magnitude rises, the time becomes increased to train BP neural. Additionally, the classification exactness shrinkages as well. A parallel design proposed to enhance the classification exactness and runtime efficiency of the BP neural network algorithm. The HFPBP and FPBP algorithms used to optimize the BP neural network’s original weights and thresholds and improve the exactness of the classification procedure. This research proposed a novel hybrid flower pollination based back propagation (HFPBP) algorithm with altered activation function and flower pollination back propagation (FPBP) algorithm with log-sigmoid activation function. The proposed HFPBP and FPBP algorithm firstly find the best search space. The investigated technique allows the convergence of global optimum resolution with more efficacy than the conventional BPNN algorithm. The proposed algorithm assessed and equated on three benchmark classification datasets, thyroid, diabetes, and glass, with conventional and proposed which is clearly shows that the proposed model performance is better than conventional with respect to low MSE (0.0005) and high accuracy (99.97%) .There are a lot of successes which are achieved by artificial neural network in various field that are such as health sciences engineering and the one field cognitive science which is unable to unnoticed due to admirable capability in intellectual complex nonlinear plotting bond and generate trainer mock-ups (Alsmadi, Omar, & Noah, 2009) .Gradient descent based back propagation algorithm are used to training of ANNs. Unfortunately, training of ANNs leading some drawbacks such as local minimum value, lower accuracy and time-consuming convergence speed Aydin (2014). To reduce these problems, the naturally enthused algorithms which are representing met heuristic algorithms are consumed to training ANNs. In illustration of research, different researchers used 2nd order stochastic learning method for the aim of training Artificial neural network mock-ups (Basheer et al., 2000). Another algorithm called Krill Herd which is also practiced for the training of ANNs (Bi et al., 2005). Bat-inspired algorithm is also used by some other researchers to adapt optimizing ANNs mock-up (Castellani & Rowlands, 2009). They used to adapt editions of bat algorithm for increasing the performance of Artificial Neural Networks. And all the given adapted structures of bat-algorithm guides to enhance performance of convergence. Metaheuristic algorithms are grouped into two techniques one is single and the other is population-based. The training of an artificial neural network which is based on a single based metaheuristic technique, it is started on one resolution, integrates with its locality to discover most excellent answer (Celik, Koylu, & Karaboga, 2016). The algorithm which is based on the Population open number of resolutions as well as produce a sequence of answers; it is not ending till it meet the condition. The algorithm based on Population similarly categorizes the evolutionary-algorithm and SI algorithm. EA include heritable algorithm that is applied in recently for a consuming Artificial Neural Network efficiency. Other researchers suggested other approaches that are giving a physics idea to instigate gravitational search technique as well biological enthused flower-pollination technique. For a suggested algorithm illustrates from the investigation outcomes, enhanced classification accuracy (Chiroma et al., 2016).
From the social conduct of animals which have a very common task consist of many trainings, the swarm intelligence algorithm inspired from that. For instance, numerous researchers used ACA for the training aim of Artificial Neural Network.
Key Terms in this Chapter
Diabetes: It is a disease that happens when the glucose of your blood, also known as blood sugar, became too high. Blood glucose is the main source of energy for every human being and it comes from the food we and you eat. Insulin are, hormones which are made by the pancreas, it helps glucose from food get into your cells and will be utilizes for energy.
Ant Colony Optimization: In the field of computer sciences and operations research, the ant colony optimization algorithm (ACO) is a probabilistic method for resolving computational issues which can be decreased to resulting best routes via graphs. Artificial Ants stand for multi-agent technique which is inspired from the actual behavior of real ants.
Artificial Neural Network: Artificial Neural Network is actually modeled is a computational model which mimic the human brains works. There are units in the ANN called neurons these units are connected to other by link and every link is associated with a weight.
Mean Square Error: In statistics the mean square error initials are (MSE) of an estimator (it is a procedure for estimating an unobserved quantity) which has the ability to measure the average of all the square of errors. It is used to verify the accuracy in the classification results.
Classification: Classification is a technique in which the data are grouped into a given number of classes on the basis of some similarity and constraints. The main aim of the classification technique is to shrink the measure of the error.
Local Minima: Local maxima are a point of a function with highest output (locally), while local minima are a point of a function with lowest output (also locally). Global maxima/minima, however, are different. They represent either the highest possible output of a function or the lowest possible output.
Optimization: It is the process in which there are various alternatives and the selection of best alternatives among them all is an optimization.
Thyroid: It is a butterfly structure gland in the inner part of the neck. It produces hormones that control the speed of metabolism in the human body and also this system helps the body to use energy. The disorders of a thyroid gland can slow down the metabolism by disrupting the manufacturing of the thyroid hormones.
Flower Pollination Algorithm: It is a bio-inspired algorithm which copies the pollination conduct of the flowering plants. Optimal plant reproduction policy includes the survival of the fittest as well as the optimal reproduction of plants in terms of numbers.
Swarm Intelligence: It is the joint conduct of dispersed, self-organized scheme, natural or artificial. The concept is used in a work on artificial intelligence. The expression was introduced by Gerardo Beni and Jing Wang in 1989, in the context of cellular robotic systems.
Accuracy: Accuracy is the parameter for evaluating the performance of the machine or models which finds that how much the measured results are near to the actual value. Accuracy calculate all the correct prediction observation divided by the total observation number or actual number.