Abstract Blood smear analysis plays an important role in the diagnosis of diseases. It is performed by doctors through visual examination of blood smears under microscope. But this is time consuming, tedious and susceptible to error. Hence automated analysis of blood smears is essential. The objective of the paper is fourfold: To determine (i) platelet count which could be used as a... Blood smear analysis plays an important role in the diagnosis of diseases. It is performed by doctors through visual examination of blood smears under microscope. But this is time consuming, tedious and susceptible to error. Hence automated analysis of blood smears is essential. The objective of the paper is fourfold: To determine (i) platelet count which could be used as a preliminary screening for dengue (ii) the type and count of White Blood Cells(WBCs) (iii) the normal and abnormal WBCs and (iv) the stage of Acute Myeloid Leukemia (AML) if abnormal from blood smear. To determine the platelet count, mapping and gray level transformation are done prior to edge detection and morphology based segmentation. To determine the type and count of WBC, the stage of AML and to determine whether WBC is normal or abnormal, threshold based segmentation is done followed by morphological operations to extract the structure of the nucleus within the blood smear. Features are extracted and further fuzzy classification is done. The experiment is conducted with real data set comprising of blood smear images from Tirunelveli Medical College Hospital (TVMCH). Besides these, images available in ASH Image Bank are also used. The results guarantee an accuracy of 100% for platelet count estimation and the classification of abnormal and normal White Blood Cells (WBC) and an accuracy of 94% for WBC type and Acute Myeloid Leukemia (AML) stage classification. The results are validated with the results of pathologist. Read More Read Less

Abstract With the advent of World Wide Web, anyone can obtain the information they want by typing in a query to the search engine. But every human has a different thought process and the results obtained from the search engine may not match their expectations. This happens due to ambiguous queries. This creates the need to personalize the search results in accordance with the... With the advent of World Wide Web, anyone can obtain the information they want by typing in a query to the search engine. But every human has a different thought process and the results obtained from the search engine may not match their expectations. This happens due to ambiguous queries. This creates the need to personalize the search results in accordance with the user preferences. To overcome the problem of ambiguous queries, we propose a new approach to collect the user preferences. In the proposed approach, clickthrough data forms the basis of capturing the user’s interest. Clickthrough data is analysed using three factors: the URL clicked, the parent domain of the URL and the category to which the clicked URL belongs to. All the three factors aid in creating a dynamic user profile which is updated using implicit feedback. To display the results, each result from search engine is assigned weight according the user profile and the results are displayed in descending order of the weights. Thus, results which interest the user are displayed at the top. We show a detailed case study of how the proposed approach works and also demonstrate its effectiveness in comparison to Google search results. Read More Read Less

Abstract Increasing demand for software made IT industries to develop high quality software within predetermined time and budget. In order to accomplish these challenges, the software development process needs to effectively managed and planned. In software development process, effort estimation is very important activity to manage and plan for effective development of software projects. If estimation of software development effort is not accurately measured... Increasing demand for software made IT industries to develop high quality software within predetermined time and budget. In order to accomplish these challenges, the software development process needs to effectively managed and planned. In software development process, effort estimation is very important activity to manage and plan for effective development of software projects. If estimation of software development effort is not accurately measured then entire software project may lead to failure and dissipate the IT industry budget. Machine learning and data mining techniques have been explored as an alternative to existing model COCOMO. This paper aims to explore artificial neural network models such as probabilistic neural networks (PNN) and generalized regression neural networks (GRNN) model on various datasets to accurately estimate the software development effort. The results are evaluated using Mean Magnitude of Relative Error (MMRE), Magnitude of Relative Error (MRE). Read More Read Less