Code smells are the indicators of the flaws in the design and development phases that decrease the maintainability and reusability of a system. A system with uneven distribution of responsibilities among the classes is generated by one of the most hazardous code smells called God Class. To address this threatening issue, an extract class refactoring technique is proposed that incorporates both cohesion and contextual aspects of a class. In this work, greater emphasis was provided on the code documentation to extract classes with higher contextual similarity. Firstly, the source code is analyzed to generate a set of cluster of extracted methods. Secondly, another set of clusters is generated by analyzing code documentation. Then, merging these two, a final cluster set is formed to extract the God Class. Finally, an automatic refactoring approach is also followed to build newly identified classes. Using two different metrics, a comparative result analysis is provided where it is shown that the cohesion among the classes is increased if the context is added in the refactoring process. Moreover, a manual inspection is conducted to ensure that the methods of the refactored classes are contextually organized. This recommendation of God Class extraction can significantly help the developers in minimizing the burden of refactoring on own their own and maintaining the software systems.

Inappropriate placement of methods causes Feature Envy (FE) code smell and makes classes coupled with each other. To achieve cohesion among classes, FE code smell can be removed using automated Move Method Refactoring (MMR) suggestions. However, challenges arise when existing techniques provide multiple MMR suggestions for a single FE instance. The developers need to manually find an appropriate target classes for applying MMR as an FE instance cannot be moved to multiple classes. In this paper, a technique is proposed named MultiMMRSReducer, to reduce multiple MMR suggestions by considering the Total Call-Frequencies of Distinct Entities (TCFDE). Experimental results show that TCFDE can reduce the multiple MMR suggestions of an FE instance and performs 77.92% better than an existing approach, namely, JDeodorant. Moreover, it can ensure minimum future changes in the dependent classes of an FE instance.

Stock market prediction is a process of trying to decide the stock trends based on the analysis of historical data. However, the stock market is subject to rapid changes. It is very difficult to predict because of its dynamic & unpredictable nature. The main goal of this paper is to present a model that can predict stock market trend. The model is implemented with the help of machine learning algorithms using eleven technical indicators. The model is trained and tested by the published stock data obtained from DSE (Dhaka Stock Exchange, Bangladesh). The empirical result reveals the effectiveness of machine learning techniques with a maximum accuracy of 86.67%, 64.13% and 69.21% for “today”, “tomorrow” and “day_after_tomorrow”.

Stock market trend can be predicted with the help of machine learning techniques. However, the stock market changes is uncertain. So it is very difficult and challenging to forecast stock price trend. The main goal of this paper is to implement a model for stock value trend prediction using share market news by machine learning techniques. Although this kind of work is implemented for the stock markets of various developed countries, it is not so common to observe such kind of analysis for the stock markets of underdeveloped countries. The model for this work is built on published stock data obtained from DSE (Dhaka Stock Exchange, Bangladesh), a representative stock market of an underdeveloped country. The empirical result reveals the effectiveness of Convolutional Neural Networks with LSTM model.