No abstract

Railways system is facing one of the biggest challenges to prevent train delay in all over the world. Categorically in India, It is far worst problem other developing countries due to increasing the number of passengers and poor update in previous system. According to the TOI newspaper, around 25.3 million people were used to travel by train in 2006 in India which drastically increased year by year[1]. In the proposed model, we used 4 different machine learning methods (linear regression, Gradient Boosting Regressor(GBR), Decision Tree and Random Forest) which have been compared with different settings to find the most accurate method. To compare different methods, we consider training time and accuracy of the method over the test data set. Trains in India get delayed frequently, and if we can predict this in advance - it would be a great help for the passengers to plan their journey according to their works. The aim of this paper is to present the prediction of Train delay in Indian Railways through machine learning techniques to achieve higher accuracy.

In recent years, mining of sequential patterns has been studied extensively in various domains. Most of the existing algorithms find patterns in transactional databases by scanning the records whether they contain patterns or not. This paper proposes a novel algorithm to mine closed sequential patterns using an inverted matrix and prefix based sequence element matrix. Inverted matrix minimizes the search space for discovering various sequential patterns of different items. We use a prefix based sequence element matrix to minimize the scans required at levels k and k+1 in the mining process. Our experimental results show the performance improvement of the new algorithm over the previous work.

A massive rise in web-based online content today pushes businesses to implement new approaches and resources that might support better navigation, processing, and handling of high-dimensional data. Over the Internet, 90% of the data is unstructured, and there are several approaches through which this data can translate into useful, structured data—classification is one such approach. Classification of knowledge into a good collection of groups is significant and necessary. As the number of machine-readable documents proliferates, automatic text classification is badly needed to classify these documents. Unlabeled documents are categorized into predefined classes of labeled documents using text labeling, a supervised learning technique. This paper reviewed some existing approaches for classifying online news articles and discusses a framework for the automatic classification of online news articles. For achieving high accuracy, different classifiers were tried. Our experimental method achieved 93% accuracy using a Bayesian classifier and present in terms of confusion metrics. ABSTRAK: Peningkatan tinggi pada masa kini pada maklumat dalam talian berasaskan web menyebabkan kaedah baru dalam bisnes telah diguna pakai dan sumber sokongan seperti navigasi, proses, dan pengurusan data berdimensi-tinggi adalah perlu. 90% data di internet adalah data tidak berstruktur, dan terdapat pelbagai kaedah data ini dapat diterjemahkan kepada data berguna, lebih berstruktur — iaitu melalui kaedah klasifikasi. Klasifikasi ilmu kepada koleksi kumpulan baik adalah penting dan perlu. Seperti mana mesin-boleh baca dokumen berkembang pesat, teks klasifikasi automatik juga sangat diperlukan bagi mengklasifikasi dokumen-dokumen ini. Dokumen yang tidak dilabel dikategori sebagai pengelasan pratakrif dokumen berlabel melalui teks label, iaitu teknik pembelajaran berpenyelia. Kajian ini mengkaji semula pendekatan sedia ada bagi artikel berita dalam talian dan membincangkan rangka kerja bagi pengelasan automatik artikel berita dalam talian. Bagi menghasilkan ketepatan yang tinggi, kami menggunakan pelbagai alat klasifikasi. Kaedah eksperimen ini mempunyai ketepatan 93% menggunakan pengelas Bayesian dan data dibentangkan berdasarkan matriks kekeliruan.

No abstract