Genetic programming for medicinal plant family identification system

Abstract

Indonesia is rich of biodiversity, but this richness is not fully utilized by the public. Due to lack of knowledge and information about medicinal plant, there are only 4.4 percent of the natural resources medicinal plants that been utilized. Difficulties in obtaining information about medicinal plants description caused people could not recognize the family or species of the medicinal plant in their environment. Generally, that information exists as text document form which is hard to be accessed. This study tries utilized critical information inside the document to identify the families of plant using heuristical approach, genetic programming. The botanical characteristic inside a document can be used to identify medicinal plant’s family, as every species has different characteristic and attributes. The Genetic Programming (GP) is able to determine the charasteristic or special feature of each family that is structurized into a tree form. The application of Genetic Programming in medicinal plant’s family identification system was aimed to assist the people in identifying medicinal plant’s family based on special appearance of each plant species on their environment. This research method consist of several phase of process, i.e. data acquisition, data booleanize, data partition into training and testing data, evaluation, and analysis. The booleanize process stated by binary numbers 1 and 0. The data as the result of booleanize process was divided into two, which are the training and testing data using 5-fold cross validation method. The training process used Genetic Programming to get the best individual, started by a generate rule or creating a number of individuals, and then those individuals was evaluated by an evaluation process called fitness evaluation. After that, genetical operation process was implemented, started by turnament process, which had an objective to select individual based on fitness value. The smaller fitness value gets the higher chances to be chosen on each generation. The crossover operation of recombining process at two individuals and development mutation operation from part of the individual was aimed to fleshing out the individual. The best individual is merely the expected solution, in the form of classification rule in identifying medicinal plants. The best classification rule with an average accuracy of 86.47% was obtained by population parameter of 10 000, and nodes of 24. Nodes consisted of function set (AND, OR, NOR) and terminal set (attribute/ special mark). In addition, the best classification rules also obtained by setting the crossover probability of 0.9, mutation probability of 0.1 and generation set to 10. The process of the training produced three rules in form of tree, which showed feature structure that differentiate each family. Those could be useful to identify the medicinal plants by public.