Soft Computing for Problem Solving 

Abstract: Text classification is used to organize documents in a predefined set of classes. It is very useful in Web content management, search engines; email filtering, etc. Text classification is a difficult task due to high- dimensional feature vector comprising noisy and irrelevant features. Various feature reduction methods have been proposed for eliminating irrelevant features as well as for reducing the dimension of feature vector. Relevant and reduced feature vector is used by machine learning model for better classification results. This paper presents various text classification approaches using machine learning techniques, and feature selection techniques for reducing the high-dimensional feature vector.

Abstract: Detecting users in an indoor environment based on Wi-Fi signal strength has a wide domain of applications. This can be used for objectives like locating users in smart home systems, locating criminals in bounded regions, obtaining the count of users on an access point etc. The paper develops an optimized model that could be deployed in monitoring and tracking devices used for locating users based on the Wi-Fi signal strength they receive in their personal devices. Here, we procure data of signal strengths from various routers, map them to the user’s location and consider this mapping as a classification problem. We train a neural network using the weights obtained by the proposed fuzzy hybrid of Particle Swarm Optimization & Gravitational Search Algorithm (FPSOGSA), an optimization strategy that results in better accuracy of the model.

Abstract: Face recognition is the most important tool in computer vision and an inevitable technology finding applications in robotics, security, and mobile devices. Though it is a technology of the past, state-of-the-art machine learning (ML) techniques have made this technology game-changing and even surpass human counterparts in terms of accuracy. This paper focuses on applying one of the advanced machine learning tools in face recognition to achieve higher accuracy. We created our own dataset and trained it on the GoogleNet (inception) deep learning model using the Caffe and Nvidia DIGITS framework. We achieved an overall accuracy of 91.43% which was fairly high enough to recognize the faces better than the conventional ML techniques. The scope of the application of deep learning is enormous and by training a huge volume of data with massive computational power, accuracy greater than 99% can be achieved. This paper will give a glimpse of deep learning, from creation of dataset to training and deploying the models, and the method can be applied for dataset corresponding to any field, be it medicine, agriculture or manufacturing, reducing the human effort and thus triggering the revolution of automation.

Abstract: Photovoltaic (PV) power generation is playing a prominent role in rural power generation systems due to its low operating and maintenance cost. The output properties of solar PV mainly depend on solar irradiation, temperature, and load impedance. Hence, the operating point of solar PV oscillates. Due to the oscillatory behavior of operating point, it is difficult to transform maximum power from the source to load. To maintain the operating point constant at the maximum power point (MPP) without oscillations, a maximum power point tracking (MPPT) technique is used. Under partial shading condition, the nonlinear characteristics of PV comprise of multiple maximum power points (MPPs). As a result, discovering true MPP is difficult. The traditional and neural network MPPT methods are not suitable to track the MPP because of oscillations around MPP and impreciseness in tracking under partial shading (PS) condition. Therefore, in this article, a biological intelligence cuckoo search optimization (CSO) technique is utilized to track and extract the maximum power of the solar PV at two PS patterns. MATLAB/Simulink is used to demonstrate the CSO MPPT operation on SEPIC converter.