TY  - JOUR
T1  - Experimental Comparison of Complexities of Artificial Neural Network and
Kolmogorov Distance for News Article Classification
AU - A. Adeyanju, Ibrahim AU - M. Olaniyan, Olatayo AU - Adeyemi, Sijuade AU - Oloyede, Ayodele AU - M. Fagbola, Temitayo AU - Omodunbi, Bolaji A. 
JO  - Journal of Engineering and Applied Sciences
VL  - 14
IS  - 7
SP  - 2276
EP  - 2291
PY  - 2019
DA  - 2001/08/19
SN  - 1816-949x
DO  - jeasci.2019.2276.2291
UR  - https://makhillpublications.co/view-article.php?doi=jeasci.2019.2276.2291
KW  - lines_of_code
KW  -neural network
KW  -Halstead measure
KW  -Kolmogorov complexity
KW  -Complexity measurement
KW  -news_article_classification
KW  -software
AB  - In today’s growing complex multi-domain and multi-language program specifications, the incessant
failure of most software products is a great challenge. This is due to the high complexities associated with these
products beyond reliable performance tolerable limits which in turn makes their maintenance impossible in an
effective manner. On one hand, the inherent complexities of software systems are often ignored at design and
implementation stages and on the other hand for such complex software products, maintenance cost becomes
very huge in the face of high defect rate. Consequently, evaluating and managing software complexities is key
to ensuring that software products are highly understandable, testable, reliable, maintainable, scalable, efficient
and cost-effective. In this study, the complexity associated with the use of Artificial Neural Network (ANN) and
Kolmogorov Complexity Distance Measure (KCDM) for solving news article classification problem was
measured using Lines of Code (LoC) and halstead measure. Similarly, the accuracy and computational efficiency
of these classifiers were also determined using true positive rate and testing time measures, respectively. British
Broadcasting Corporation (BBC) News dataset composed of 2225 documents corresponding to entertainment,
sport, education/technology, politics and business was used for experimental purpose. Based on the results
obtained, ANN produced higher classification accuracy at higher complexity and classification time while
KCDM yielded lower complexity and testing time but suffers from low classification accuracy. Hence, from a
developer’s point of view, complexity measurement at design and implementation stages of the software
development life cycle is pertinent in other to identify complex designs and codes for refactoring to prevent
high software defect rate. From the stakeholders end, effective decisions can be reached by considering the
tradeoffs between complexity and accuracy of software products to be used in real-life settings.
ER  -