Dynamic Behavior Extraction from Social Interactions Using Machine Learning and Study of Over Fitting Problem

People in this world interact with each other directly (face-toface) or indirectly (by sending text messages). As humans are linked with emotions themselves it is easy for them to detect emotions but same is not true with computers. But for a computer detecting an emotion will be a difficult job to perform. Emotion is cognitive process where machine learning techniques are used to detect it. The main goal of the machine learning algorithms is to build computer system that can adopt and learn from their experience or by examples. This paper presents a systematic performance analysis of four classification algorithms for the extraction of human emotions and does also over fitting problem analysis. Our main attempt in this paper is to detect emotions from social interactions on twitter social network. Four machine learning algorithm are used for text classification under seven emotional classes. The work is carried on two data bases one data base is build by collecting live tweets from twitter social network site and second ISEAR (International Survey on Emotion Antecedents and Reactions) database. WEKA interface is used for the implementation of our work which shows above 85% accuracy in identifying the emotion classes. The ranking and standard deviation functionalities provided by the WEKA experimenter helps to determine the effectiveness of a classifier model.