|
|
| |
| Articles
by
S.A. Noah |
Total Records (
4 ) for
S.A. Noah |
|
 |
| |
|
| |
S.A. Noah
and
F. Ismail
|
| |
This research presented an experimental study on automatic
classification of Malay proverbs using Naïve Bayesian algorithm.
The automatic classification tasks were implemented using two Bayesian
models, multinomial and multivariate Bernoulli model. Both models were
calibrated using one thousand training and testing dataset which were
classified into five categories: family, life, destiny, social and knowledge.
Two types of testing have been conducted; testing on dataset with stop
words and dataset with no stop words by using three cases of Malay proverbs,
i.e., proverb alone, proverb with meaning and proverb with the meaning
and example sentences. The intuition was that, since proverbs were commonly
short statement, the inclusion of its meaning and associated used in sentences
could improve the accuracy of classification. The results showed that
a maximum of 72.2 and 68.2% of accuracy have been achieved respectively
by the Multinomial model and the Multivariate Bernoulli for the dataset
with no stop words using proverb with the meaning and example sentences.
This experiment has indicated the capability of the Naïve Bayesian
algorithm in performing proverbs classification particularly with the
inclusion of meaning and example usage of such proverbs. |
|
| |
|
| |
M. Alsmadi
,
K.B. Omar
,
S.A. Noah
and
I. Almarashdeh
|
| |
The aim of this study is a novel hybrid approach for optimizing the performance of back-propagation classifier (BPC) by utilizing the ability of Memetic algorithm (genetic algorithm and great deluge algorithm) to optimize the parameters (weight) of the PBC for fish classification problem. To recognize an isolated pattern of fish in the image based on the combination between robust features extraction which extracted based on Potential Local Geometric Features (PLGF) and shape measurements, which are extracted by measuring the edge detection method, distance and angle measurements. Typical the BPC has such disadvantage as slow practice speed and easy for running into local minimum. We presented a system prototype for dealing with such problem. The process started by acquiring an image-containing pattern of fish, then the image features extraction is performed relying on PLGF and shape measurements. The hybrid Memetic Algorithm (genetic algorithm and great deluge algorithm) with BPC (HGAGD-BPC) has outperformed BPC method and previous methodologies by obtaining better quality results but with a high cost of computational time compared to the BPC method. Where the overall accuracy obtained using the traditional BPC was 86%, while the overall accuracy obtained by the HGAGD-BPC was 96% on the test dataset used. We developed a classifier for fish images classification. Eventually, the classifier is able to classify the given fish into poison and non-poison fish and classify the poison and non-poison fish into its family. |
|
|
| |
|
| |
I. Zaharudin
,
S.A. Noah
and
M.M. Noor
|
| |
In this study a semi automatic acquisition of domain
relevant terms from digital documents in e-newspaper related to Malaysian
medicinal herbs is presented. This study proposes (1) TFIDF-based term
classification method for acquiring single word terms, (2) recognition
of multi-word using TerMine software to acquire multiword terms and (3)
Hearst`s methodology of acquiring semantic relationships of hyponym. The
results show the benefits of using these methods in selecting relevant
terms from domain specific corpus. From this study it is believed that
the combination of these three methods might be helpful to select relevant
terms as well as minimize the effort to discard irrelevant terms manually
from wide collection of terms from the corpus. |
|
|
| |
|
| |
S.A. Noah
,
A. Azilawati
,
T.M. Tengku Sembok
and
T.W. Tengku Siti Meriam
|
| |
Web documents contain useful textual information that can be exploited for describing images. Research had been focused on representing images by means of its content (low level) description such as color, shape and texture, little research had been directed to exploiting such textual information. The aim of this research was to systematically exploit the textual content of HTML documents for automatically indexing and ranking of images embedded in web documents. A heuristic approach for locating and assigning weight surrounding web images and a modified tf.idf weighting scheme was proposed. Precision-recall measures of evaluation had been conducted for ten queries and promising results had been achieved. The proposed approach showed slightly better precision measure as compared to a popular search engine with an average of 0.63 and 0.55 relative precision measures respectively. |
|
|
|
|
| |
|
