Nature-Inspired Metaheuristic Support Vector Classification System for Enhanced Prediction in Geotechnical Engineering
| Autor: | Julian Pratama Putra Thedja, 鄭天洪 |
|---|---|
| Rok vydání: | 2015 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 103 Advanced data mining techniques are potential tools for solving geotechnical engineering problems. This study proposes a novel classification system integrating swarm and evolutionary intelligence, i.e., a smart firefly algorithm (SFA), with a least squares support vector machine (LSSVM) algorithm. The SFA is an optimization algorithm that involves combining a firefly algorithm (FA) with metaheuristic components, namely chaotic logistic map, chaotic Gauss/mouse map, adaptive inertia weight, and Lévy flight, to enhance the performance of the FA in optimization problems. Nine benchmark functions were used to validate the performance of the SFA. The experimental results showed that the SFA solved seven benchmark functions at a 100% success rate and that it solved two benchmark functions at an 84%–87% success rate. The LSSVM algorithm was adopted in this study because of its excellent performance in solving two-spiral classification problems, which are difficult to solve using multilayer perceptrons. The SFA was then integrated with the LSSVM algorithm to create a hybrid system called SFA-LSSVM to automatically tune LSSVM hyperparameters for enhancing the LSSVM performance. A graphical user interface was developed for the proposed classification system to assist engineers and researchers in executing advanced data mining tasks. The performance of the proposed system was compared with that of those reported in previous works by using a cross-validation algorithm. The system was applied to several case studies that involved measuring the groutability of sandy silt soil, monitoring seismic hazards in coal mines, predicting postearthquake soil liquefaction, and determining risk preference in slope collapse. These case studies involved geotechnical engineering problems that can lead to disastrous consequences. The prediction problems in these studies were complex because they were dependent on various physical factors, and such factors exhibited highly nonlinear relations. The results revealed that the proposed SFA-LSSVM system exhibited a groutability prediction accuracy of 95.41%, seismic prediction accuracy of 93.96%, soil liquefaction prediction accuracy of 95.18%, and soil collapse prediction accuracy of 95.45%. Hence, the proposed system is a promising tool to help decision-makers in geotechnical engineering planning and design tasks. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|