A study of feature selection algorithms for automated mining of medical tweets

Abstract

Mining opinions from micro blogging sites like twitter helps us in decision making in the platforms like business, shopping, medical, politics, etc. Twitter sentiment analysis helps us in examining various real world problems. There are many challenges in handling the medical tweets which led to many proposed techniques to achieve the best accuracy in sentiment analysis. Hence in this research many nature inspired algorithms are said to be tested for increasing the accuracy in opinion mining of medical tweets. Priory cancer and drug data set has been tested using the heuristic Genetic algorithm. In which roulette wheel selection, two point crossover and flip bit mutation have been applied. Since the simple Genetic algorithm performs only local search the research is extended for metaheuristic algorithms which tests for global best solutions. And many nature inspired algorithms have been tested for sample data. By the first work of research the exploration and exploitation of Shuffled Frog Leaping algorithm (SFLA) are improved to develop enhanced feature selection technique using Modified Shuffled Frog Leaping algorithm (MSFLA). In SFLA the population of frogs is grouped in to memeplexes and the local search operation is performed for several iterations. New memeplexes are created by shuffling the memeplexes for the best convergence rate. The convergence speed of SFLA decreases because of updating worst solutions instead of best solutions. The limitations in existing algorithm SFLA is overcome by MSFLA by applying the crossover for improving the worst positions of frogs. The parental features are obtained by applying the crossover between worst and best ranked features newline