Computational Intelligence Based Machine Learning Methods For Rule-Based Reasoning In Computer Vision Applications T T Dhivyaprabha 1 , Research Scholar P Subashini 1 , Professor M Krishnaveni 2 , Assistant Professor Department of Computer Science Avinashilingam Institute for Home Science and Higher Education for Women Coimbatore, India ttdhivyaprabha@gmail.com mail.p.subashini@gmail.com krishnaveni.rd@gmail.com Abstract— In robot control, rule discovery for understanding of data is of critical importance. Basically, understanding of data depends upon logical rules, similarity evaluation and graphical methods. The expert system collects training examples separately by exploring an anonymous environment by using machine learning techniques. In dynamic environments, future actions are determined by sequences of perceptions thus encoded as rule base. This paper is focused on demonstrating the extraction and application of logical rules for image understanding, using newly developed Synergistic Fibroblast Optimization (SFO) algorithm with well-known existing artificial learning methods. The SFO algorithm is tested in two modes: Michigan and Pittsburgh approach. Optimal rule discovery is evaluated by describing continuous data and verifying accuracy and error level at optimization phase. In this work, Monk’s problem is solved by discovering optimal rules that enhance the generalization and comprehensibility of a robot classification system in classifying the objects from extracted attributes to effectively categorize its domain. Keywords— rule discovery; machine learning; Synergistic Fibroblast Optimization (SFO) algorithm; classification problem; robot systems; I. INTRODUCTION Basically, the idea of artificial intelligence is to replicate the human ways of reasoning in computing. A system can be defined as intelligent, only if it satisfies learning and decision making requirements. Many scientific problems are solved by numerical solutions with high performance computing, where other problems do not have definite algorithms which demand a need of effective algorithms that give solutions through intelligence. Even non-algorithm based problems utilize the field of computational intelligence in areas such as perception, visual perception, control and planning problems in robotics and in many non-linear complex problems. In general, artificial intelligence deals with soft computing that create an interface with reality, whereas intelligence system provides solutions in handling practical computing problems. The knowledge based system (KBS) consists of rule-based reasoning (RBR), case- based reasoning (CBR) and model-based reasoning (MBR), where the combination of KBS can be CBR–RBR, CBR– MBR and RBR–CBR–MBR, and also Intelligent Computing method has the combination of ANN–GA, fuzzy–ANN, fuzzy–GA and fuzzy–ANN–GA. Though there are many methods available for combining soft and hard computing, this paper presents a hybrid method which is effectively carried out by fusing software and hardware computing interconnected with an application. In this research work, a crucial work, based on monk’s problem is taken to solve the classification problems, that would be more useful for the field of Human-Computer Interaction (HCI) which has the application of vision and speech analysis. The proposed hybrid model is developed based on a new optimization technique, Synergistic Fibroblast Optimization (SFO), by applying with some well known machine learning methods to solve monk’s classification problem. The implemented algorithm obtains optimal logical rules that classify the dataset into predicted class labels. It had been tested on both single mode and iterative mode, where the accuracy of the classification model mainly depends on the quality of knowledge acquired, which can be represented in terms of set of rules. The general description for defining rules is shown in Equation 1, where the rule consists of feature variables (attribute values) and target variable (class). IF <attribute 1 relational_operator value 1 logical_condition attribute n relational_operator value n > THEN<Class> (1) The implemented SFO discovers certain potential sets of IF-THEN classification rules encoded into real-valued collagens that contain all types of attribute values and corresponding positive and negative classes in dataset. This implication finds optimal solutions that solve error minimisation problem, and the examined results have indicated that SFO outperforms original PSO algorithm [10]. This work is supported by the project titled “Design and Development of Computer Assistive Technology System for Differently Abled Student” (No. SB/EMEQ-152/2013) funded by Science and Engineering Research Board (SERB)-Department of Science & Technology (DST).