Article Preview
Top1. Introduction
Data mining used in knowledge discovery (Hussein et al., 2019), like supervised and unsupervised machine learning techniques (Willis and Strunk, 2017), helps in the decision-making process. Supervised machine learning, like classification (Michael and Constantin, 2002) and the derivative regression, is used in the efficient categorization of data that is based on a set of rules and perspectives. Data must be processed before using it in the classification technique (Kumar et al., 2019).
Prediction is considered a major application of supervised machine learning algorithms (Zaho et al., 2016) and depends mainly on well-optimized parameters concerning supervised data mining algorithm. The applicability of prediction models could be found in electricity price estimation (Shrivastava & Khosravi., 2014) that helps in managing uncertainty and fraud that could be found in electricity prices. Unsupervised machine learning techniques, like clustering (Fouad and Dawood, 2016), aims to divide the dataset into partitions called clusters.
The swarm intelligence principle (Eberhart et al., 2001) was presented in the intelligence computation domain in (Beni and Wang, 1989), which was inspired by the activities in neurosciences and behavioral sciences, as an intelligent paradigm to handle issues, especially in the optimization domain, without a global model provision. A swarm (Ab-Wahab et al., 2015) is a population of identical, agents accomplishing incipient tasks and reacting among themselves, and their ambiance, without lacking central domination. In this situation, particle swarm optimization PSO, primarily presented by (Kennedy and Eberhart, 1995, Blum and Merkle, 2008), is a meta-heuristic global optimization method, which is related to the tribe of algorithms based on the swarm intelligence concept. The PSO technique, which was presented by Kennedy and Eberhart (Kennedy et al., 1995), is a metaheuristic algorithm assorted in swarm intelligence methodologies. PSO emulates the harmonious disposal of flocks moving for birds and fish that convey information across the group to assist the decision-making process in a synchronized way. While the flocks are moving to discover food, each particle detects its position and speed.
In this paper, the enhanced prediction through scalable data mining is performed by proposing an intelligent approach. The proposed approach aims at integrating the enhanced PSO with the SVR to provide an effective prediction process. The enhanced PSO, called PLTVACIW-PSO, is based on Parallelized Linear Time-Variant Acceleration Coefficients (TVAC) and Inertia Weight (IW) of PSO. The proposed optimization algorithm, PLTVACIW-PSO, is exploited to optimize and adjust SVR parameters, PLTVACIW-PSO-SVR. PLTVACIW-PSO-SVR is evaluated by performing the experimental comparisons of the proposed algorithm with eleven different algorithms, which are shown in Table 1. The experimental comparisons are achieved by applying the proposed algorithm and these algorithms to twenty-one different datasets varying in their scales. Furthermore, the evaluation considers the execution time to prove that PLTVACIW-PSO-SVR is performed efficiently on different datasets compared with the other eleven algorithms.
Table 1. No. | Algorithm Name | Short Name | Category | Experiment Phase | Referenced or Implemented | Description |
1 | Linear Inertia Weight PSO | LIW-PSO | IW Family | 1 and 2 | Zheng et al., 2003 | Inertia weight w is varying overtime linearly. |
2 | Parallelized Linear Inertia Weight PSO | PLIW-PSO | 1 and 2 | implemented | A parallelized version of LIW-PSO |
3 | Non-Linear Inertia Weight PSO | NLIW-PSO | 1 and 2 | Chatterjee et al., 2006 | Inertia weight w is varying non-linearly overtime. |
4 | Parallelized Non-Linear Inertia Weight PSO | PNLIW-PSO | 1 and 2 | implemented | A parallelized version of NLIW-PSO |
5 | Linear Time-Variant PSO | LTV-PSO | TVAC Family | 1 and 2 | Ratnaweera et al., 2004 | PSO acceleration variables c1 and c2 are varying overtime linearly. |
6 | Parallelized Linear Time-Variant PSO | PLTV-PSO | 1 and 2 | Fouad et al., 2017 | A parallelized version of LTV-PSO |
7 | Non-Linear Time-Variant PSO | NLTV-PSO | 1 and 2 | Chen et al., 2009 | PSO acceleration variables c1 and c2 are varying non-linearly overtime. |
8 | Parallelized Non-Linear Time-Variant PSO | PNLTV-PSO | 1 and 2 | Fouad et al., 2017 | A parallelized version of NLTV-PSO |
9 | Particle Swarm Optimization | PSO | Traditional Algorithms | 1 and 2 | Kennedy and Eberhart., 1995 | An algorithm is inspired by the behavior of bird swarms and fish. |
10 | Differential Evolution | DE | 1 and 2 | Price et al., 2006 | A modified version of genetic algorithms |
11 | Genetic Algorithms | GA | 1 and 2 | Coley., 1999 | A natural-inspired optimization algorithm based on human gens and its cross over and mutation |