Cloud Theory And Its Application In Plant Numerical Taxonomy Of Genus Camellia

A new cloud-fuzzy based classifier has been proposed in this paper.The proposed PSOCCAS(particle swarm optimization aided cloud classifier based on attribute similarities) is extended from the cloud model and particle swarm optimization algorithm.The performance of the proposed PSOCCAS is well demonstrated by employing it for a benchmark problems and comparing them with several other algorithms available in practice.The testing accuracy is found very encouraging.The performance of our proposed system is only bettered by some GA or EA-based fuzzy systems which showed fantastic results.For the purpose of applying the PSOCCAS in practical use,the datasets of genus Camellia and Fisher's iris data is obtained from the famous UCI machine learning repository(The iris data set presents a four dimensional input problem with 150 instances,sepal length,sepal width,petal length and petal width as the input attributes, and Setosa,Versicolor and Virginica as the three possible flowers in which the output can be classified),are analyzed in this paper.There are five main outstretched functions developed.First of all,to highlight the excellence of the proposed algorithms,we compared its accuracy with several well-known classifiers,employed before for the same problem.Some of these algorithms employed classical approaches while the others employed soft-computing based methodologies.Results showed that the performance of the PSOCCAS is among(Classification performance:Versicolor 96%,Virginica 98%,Setosa 100%and average of 98%) the best ones achieved.Some of these algorithms employed fuzzy models,which were developed employing genetic algorithm(GA) and evolutionary algorithm(EA)-based techniques.They included both Mamdani-type and Takagi-Sugeno type fuzzy models.Among them,it can be seen that those algorithms described in Setnes and Roubos,Chang and Lilly showed excellent performance results.However,it should be noted that GA and EA-based approaches are in existence for quite some time now and these fields have become quite rich over the past couple of decades.All these algorithms in Setnes and Roubos, and Chang and Lilly employ efficient,improved variants of evolutionary optimization.Secondly,we have demonstrated by experiments that the taxonomic results based on all available features have shown the superiority performance over the ones based on portions of attributes.For example,the model which is based on features with small weights(MSW)—combined with LAE,LPM and LVL,get higher misplacing rate.In this cloud model,45.83%of species belonged to Sect.Furfuracea are placed into Sect.Camellia's model(S2(x)).Comparatively,the model based on features (combined with ABA,ASP and LSI) with large weights(MLW) has higher accuracy rate of classification,with only 25.00%of species belonged to Sect.Furfuracea are misplaced in Sect.Camellia's model(S2(x)).These results could describe the reason why all these algorithms could not easy get the 100%accurate classification results. Accordingly,it can clearly describe the phenomenon in plant numerical taxonomy that different taxonomic treatments based on different portions of features generate not the same result and explain why divergences exist in different taxonomic systems.Thirdly,the weights of the considered features will do much help in establishing a validly plant Key in the Flora.Since features with large weights contribute much more to the classification,priorities of consideration are suggested to be assigned to them.Take Sect.Furfuracea for example,the character of leaf margin,amount of the petals and leaf area,etc.,are preferentially used in the existent plant Key of Flora of China.However,we suggest the aforementioned five features,especially the ABA, ASP and FDM by sort descending,be significantly considered in the plant Flora Key. On the other hand,the weight of the same feature may be different in different taxa. But if the weights of the considered features are measured firstly,the plant Flora Key should be handy and scientific without any subjective errors.What's more,the expected values of the considered features could best describe the main character of taxa.Take Sect.Furfuracea for example,the above-mentioned five features of this section(ABA,ASP,FDM,TSP and LSI) are 14,8.4 cm,334.6 um,2.9 cm, respectively.However,those features with small weights are not uselessness in an objective and scientific taxonomic treatment.When a taxonomy treatment is to be established,their effects should be validly considered.Fourthly,we applying the proposed classifier for selecting the 'expected species', which has perfect actions in taxonomy been proved.We find some interesting overlaps between the 'type species' and 'expected species':the selected expected species C.furfuracea and C.japoncia are also type species of Sect.Furfuracea and Sect.Camellia,respectively.It implies that the selected expected species can indicate their sections,too.In other words,the PSOCCAS could be a good tool for solving the divergence within section level in the numerical taxonomy of genus Camellia.Here we suggest that the expected species be served as an illustration in plant numerical taxonomy.Though it seems to be similar with the type species in traditional taxonomy, many differences between them are very significant.Sometimes,they're not the same species in one section.Take Sect.Thea for example,the expected species selected by the algorithm is C.jingyunshanica,but the type species is C.sinensis.Our purpose is to dig the superiority of its expected characters in numerical taxonomy.Then,the ascription of a divergent(or new) species is no more an endless divergences since its membership degree could be validly calculated based on the PSOCCAS and the new illustration.Last but not least,we used the PsoccAs method to evaluate the species sources of Camellia plants(Sect.Chrysantha Chang,Sect.Archecamellia Sealy,Sect. Oleifera Chang,Sect.Stereocarpus(Pierre) Sealy,Sect.Furfuracea Chang and Sect. Luteoflora Chang) based on their photosynthesis and chlorophyll fluorescence characters.As these characters present much inner relationships between plants during their evolution with the changing entironment,these characters could be significant bases in plant phylogenetic and systematics.On the other hand,these characters are useful information for cultivated culture.Based on the characters(with great weight values) of Ca(weight=0.4916),RCOC(weight=0.2146) and E(weight=0.1740), the studied six sections were clustered into three eco-types:all species expect C. vietnamensis Huang ex Hu in Sect.Oleifera Chang were classified as TypeⅠ,which have highest photosynthesis capability and distribute under high intensive light areas. Sect.Luteoflora Chang was classified as TypeⅡ,this eco-type has lowest photosynthesis activity,and distributes strictly under low intensive light area.C. vietnamensis Huang ex Hu and Sect.Chrysantha Chang,Sect.Archecamellia Sealy, Sect.Stereocarpus(Pierre) Sealy,Sect.Furfuracea Chang were classified as TypeⅢ, this species of this type have a wide distributions.Accordingly,the PSOCCAS will be a landmark method for solving many divergences in plant numerical taxonomy.