Search

Nowadays, embedded linux is widely used operating system due to its compatibility with variety of systems like super computer, work station, personal computer, system on chip (SoC), server and custom computer. The major reason of using linux is high portability and open source code. Also, it supports different type of architectures like- x86, PowerPC, MIPS, H8, SPARC, or ARM and many more. This paper deals with design, implementation and testing of character device driver for the GPIO pins of ARM Cortex based platform. Device driver is a piece of code written in C language which is responsible for controlling the hardware and part of the kernel. ARM based platform Raspberry Pi is used in this work, supports the Linux OS. Due to missing of dependencies and patches, the option of cross compilation is available such as tool chain which contains GCC compiler, assembler, linker and debugger etc.

The Higher Secondary Certificate (HSC) Examination in Bangladesh was very crucial for students' future, as well as job lives. As the performance of this examination was influenced by some factors, the main scope of this research was to identify these influencing factors with highest possible precision. In this research, an experimental methodology was adopted to generate a dataset which was constructed from students who passed HSC. Several Data mining methods were used to analyze data, extracting models, and develop knowledge to make decision and also used some feature selection methods to remove unnecessary and redundant features which were helped to improve the performance of the classifiers. Here applied several existent feature selection algorithms and a proposed approach to find the relevant features. The results of the empirical analysis were obtained by removing these irrelevant features with some bayes based, function based, lazy based, rule base and tree base classifiers. And finally the results showed, feature reduction had some effect on performance and for overall LMT classifier with our selected features performed better than others.

To ignore noisy, skewed, correlated, imbalanced and unnecessary features from real life problems, the feature subset selection with learning algorithm was faced some problems of selecting these relevant features. Several factors like-skewed, high kurtosis valued, dependence or correlation influenced features as well as the classifiers performance. A Dropout technique (feature subset selection method) was used to select relevant and important features and mixed with a classifier named as Probabilistic Neural Network (PNN) to find the performance. This study conducted Dropout technique with PNN to prevent above irrelevant features to get better performance. The main task of this method was to apply several techniques (by omitting some fixed features like-skewed, high kurtosis valued, dependent or correlated and Dropout) with PNN in several datasets and noticed that the overall performance for all datasets was significantly improved or unchanged for Dropout with PNN(DPNN) and DPNN always performed better than others.