Nailah Al-Madi

Training radial basis function networks using biogeography-based optimizer

Training artificial neural networks is considered as one of the most challenging machine learning problems. This is mainly due to the presence of a large number of solutions and changes in the search space for different datasets. Conventional training techniques mostly suffer from local optima stagnation and degraded convergence, which make them impractical for datasets with many features. The literature shows that stochastic population-based optimization techniques suit this problem better and are reliably alternative because of high local optima avoidance and flexibility. For the first time, this work proposes a new learning mechanism for radial basis function networks based on biogeography-based optimizer as one of the most well-regarded optimizers in the literature. To prove the efficacy of the proposed methodology, it is employed to solve 12 well-known datasets and compared to 11 current training algorithms including gradient-based and stochastic approaches. The paper considers changing the number of neurons and investigating the performance of algorithms on radial basis function networks with different number of parameters as well. A statistical test is also conducted to judge about the significance of the results. The results show that the biogeography-based optimizer trainer is able to substantially outperform the current training algorithms on all datasets in terms of classification accuracy, speed of convergence, and entrapment in local optima. In addition, the comparison of trainers on radial basis function networks with different neurons size reveal that the biogeography-based optimizer trainer is able to train radial basis function networks with different number of structural parameters effectively.

Parallel glowworm swarm optimization clustering algorithm based on MapReduce

Clustering large data is one of the recently challenging tasks that is used in many application areas such as social networking, bioinformatics and many others. Traditional clustering algorithms need to be modified to handle the increasing data sizes. In this paper, a scalable design and implementation of glowworm swarm optimization clustering (MRCGSO) using MapReduce is introduced to handle big data. The proposed algorithm uses glowworm swarm optimization to formulate the clustering algorithm. Glowworm swarm optimization is used to take advantage of its ability in solving multimodal problems, which in terms of clustering means finding multiple centroids. MRCGSO uses the MapReduce methodology for the parallelization since it provides fault tolerance, load balancing and data locality. The experimental results reveal that MRCGSO scales very well with increasing data set sizes and achieves a very close to linear speedup while maintaining the clustering quality.

Optimizing the Learning Process of Feedforward Neural Networks Using Lightning Search Algorithm

Evolutionary Neural Networks are proven to be beneficial in solving challenging datasets mainly due to the high local optima avoidance. Stochastic operators in such techniques reduce the probability of stagnation in local solutions and assist them to supersede conventional training algorithms such as Back Propagation (BP) and Levenberg-Marquardt (LM). According to the No-Free-Lunch (NFL), however, there is no optimization technique for solving all optimization problems. This means that a Neural Network trained by a new algorithm has the potential to solve a new set of problems or outperform the current techniques in solving existing problems. This motivates our attempts to investigate the efficiency of the recently proposed Evolutionary Algorithm called Lightning Search Algorithm (LSA) in training Neural Network for the first time in the literature. The LSA-based trainer is benchmarked on 16 popular medical diagnosis problems and compared to BP, LM, and 6 other evolutionary trainers. The quantitative and qualitative results show that the LSA algorithm is able to show not only better local solutions avoidance but also faster convergence speed compared to the other algorithms employed. In addition, the statistical test conducted proves that the LSA-based trainer is significantly superior in comparison with the current algorithms on the majority of datasets.

Adaptive genetic programming applied to classification in data mining

Classification is a data mining method that assigns items in a collection to target classes with the goal to accurately predict the target class for each item in the data. Genetic programming (GP) is one of the effective evolutionary computation techniques to solve classification problems, however, it suffers from a long run time. In addition, there are many parameters that need to be set before the GP is run. In this paper, we propose an adaptive GP that automatically determines the best parameters of a run, and executes the classification faster than standard GP. This adaptive GP has three variations. The first variant consists of an adaptive selection process ensuring that the produced solutions in the next generation are better than the solutions in the previous generation. The second variant adapts the crossover and mutation rates by modifying the probabilities ensuring that a solution with a high fitness is protected. And the third variant is an adaptive function list that automatically changes the functions used by deleting the functions that do not favorably contribute to the classification. These proposed variations were implemented and compared to the standard GP. The results show that a significant speedup can be achieved by obtaining similar classification accuracies.

Improving genetic programming classification for binary and multiclass datasets

Genetic Programming (GP) is one of the evolutionary computation techniques that is used for the classification process. GP has shown that good accuracy values especially for binary classifications can be achieved, however, for multiclass classification unfortunately GP does not obtain high accuracy results. In this paper, we propose two approaches in order to improve the GP classification task. One approach (GP-K) uses the K-means clustering technique in order to transform the produced value of GP into class labels. The second approach (GP-D) uses a discretization technique to perform the transformation. A comparison of the original GP, GP-K and GP-D was conducted using binary and multiclass datasets. In addition, a comparison with other state-of-the-art classifiers was performed. The results reveal that GP-K shows good improvement in terms of accuracy compared to the original GP, however, it has a slightly longer execution time. GP-D also achieves higher accuracy values than the original GP as well as GP-K, and the comparison with the state-of-the-art classifiers reveal competitive accuracy values.

Scaling Genetic Programming for data classification using MapReduce methodology

Genetic Programming (GP) is an optimization method that has proved to achieve good results. It solves problems by generating programs and applying natural operations on these programs until a good solution is found. GP has been used to solve many classifications problems, however, its drawback is the long execution time. When GP is applied on the classification task, the execution time proportionally increases with the dataset size. Therefore, to manage the long execution time, the GP algorithm is parallelized in order to speed up the classification process. Our GP is implemented based on the MapReduce methodology (abbreviated as MRGP), in order to benefit from the MapReduce concept in terms of fault tolerance, load balancing, and data locality. MRGP does not only accelerate the execution time of GP for large datasets, it also provides the ability to use large population sizes, thus finding the best result in fewer numbers of generations. MRGP is evaluated using different population sizes ranging from 1,000 to 100,000 measuring the accuracy, scalability, and speedup.

PSO-AG: A Multi-Robot Path Planning and obstacle avoidance algorithm

Robot path planning is one of the most challenging tasks as it involves several parameters and different constraints. Robots moving in an environment face many challenges such as avoiding obstacles. Path planning aims at directing the robot to reach a target via a collision-free path. Online Path Planning allows robots to move in an environment they do not have prior knowledge about and ought to discover while moving. This paper introduces, PSO-AG, an online multi robot path planning algorithm that combines the benefits of particle swarm optimization and Agoraphilic algorithms. In PSO-AG, particle swarm optimization works as the moving path planner that decides the next point for the robots to reach the target, and Agoraphilic works as the moving controller that steers the robots towards the target while avoiding obstacles along the path. Simulation was used to evaluate the performance of PSO-AG in different scenarios; including different sizes of robots swarms and different levels of environment difficulty; ranging from obstacle-free to partially obstructed environment. Experiments showed promising results of PSO-AGs scalability and target reaching rate.

Segment-based genetic programming

Genetic Programming (GP) is one of the successful evolutionary computation techniques applied to solve classification problems, by searching for the best classification model applying the fitness evaluation. The fitness evaluation process greatly impacts the overall execution time of GP and is therefore the focus of this research study. This paper proposes a segment-based GP (SegGP) technique that reduces the execution time of GP by partitioning the dataset into segments, and using the segments in the fitness evaluation process. Experiments were done using four datasets and the results show that SegGP can obtain higher or similar accuracy results in shorter execution time compared to standard GP.

The Influence of Input Data Standardization Methods on the Prediction Accuracy of Genetic Programming Generated Classifiers

Genetic programming (GP) is a powerful classification technique. It is interpretable and it can dynamically build very complex expressions that maximize or minimize some fitness functions. It has a capacity to model very complex problems in the area of Machine Learning, Data Mining and Pattern Recognition. Nevertheless, GP has a high computational complexity time. On the other side, data standardization is one of the most important pre-processing steps in machine learning. The purpose of this step is to unify the scale of all input features to have equal contribution to the model. The objective of this paper is to investigate the influence of input data standardization methods on GP, and how it affects its prediction accuracy. Six different methods of input data standardization were checked in order to determine which one allows to achieve the most accurate result with lowest computational cost. The simulations have been implemented on ten benchmarked datasets with three different scenarios (varying the population size and number of generations). The results showed that the computational efficiency of GP is highly enhanced when coupled with some standardization methods, specifically Min-Max method for scenario I and Vector method for scenario II, and scenario III. Whereas, Manhattan and Z-Score methods had the worst results for all three scenarios.

Energy-aware geographic routing protocol with sleep scheduling for wireless multimedia sensor networks

In Wireless Multimedia Sensor Networks, the strict quality-of-service requirements mandate discovering and routing via shortest paths. The intensity of the multimedia traffic causes relay nodes energy to deplete relatively fast. This often leads to breaking the routing path and causing the transmission to halt until a new path is discovered. Keeping the inactive relay nodes awake all the time causes their energy to deplete uselessly and decreases the chance of discovering alternative paths. This paper introduces a geographic routing protocol called (EA-TPGF-SS) with energy-awareness, simple synchronized sleep scheduling mechanism, and feedback capability on the residual energy of the relay nodes. The performance evaluation and comparative analysis show that the proposed approach can significantly improve the network lifetime, end-to-end delay, number of delivered packets, and delivered packets per Joule; compared to existing similar approaches, however it has a slight decrease in the overall throughput.