Browsing by Author "Amusa, Lateef"

Now showing 1 - 2 of 2
  • Item
    DATA MINING OF NIGERIANS’SENTIMENTS ON THE ADMINISTRATION OF FEDERAL GOVERNMENT OF NIGERIA
    (Computers and Applied Computer Science Faculty in "Tibiscus" University of Timişoara, Romania., 2016) Amusa, Lateef; Yahya, Wahab; Balogun, Abdullateef Oluwagbemiga
    The opinions and sentiments expressed by citizens of a country on the policies of the government of such country are very vital to the overall running of the affairs of such a government. This paper therefore explored data mining tools to evaluate peoples’ sentiments (positive or negative) towards the administration of the Federal Government of Nigeria (FGN) under President Muhammadu Buhari (PMB). Data were collected through a popular social medial network (Twitter) on various tweets by Nigerians with respect to their perceptions about the current administration PMB. The simple but powerful Naïve Bayes (NB) classifier was adopted to classify the various tweets submitted by Nigerians through this medium into positive and negative sentiments. For polarity, it was trained on the combination of Janyce Wiebe’s subjectivity lexicon and Bing Liu’s subjectivity lexicon which polarized the submitted words as being negative or positive. Out of about 13,000 features (peoples’ sentiments) considered, 4,770 of them were used after data cleaning. The results showed that the proportion of positive and negative sentiments, as obtained from the data, were 45.2% and 54.8% respectively. However, the data were randomly partitioned into 80:20 training and testing parts respectively and the NB classifier was learned on the training set while its goodness was assessed on the test set. The prediction accuracy, misclassification error rate, sensitivity and specificity of the classifier were 78.3%, 21.7%, 82.5% and 88.1% respectively. All analyses were carried out in the environment of R statistical package (version 3.2.2).
  • Item
    An Ensemble Approach Based on Decision Tree and Bayesian Network for Intrusion Detection
    (Computers and Applied Computer Science Faculty in "Tibiscus" University of Timişoara, Romania., 2017-06-01) Balogun, Abdullateef Oluwagbemiga; Balogun, Adedayo Miftaudeen; Sadiku, Peter Ogirima; Amusa, Lateef
    This paper presents an overview of intrusion detection and a hybrid classification algorithm based on ensemble method (stacking) which uses decision tree (J48) and Bayesian network as base classifiers and functional tree algorithm as the meta-learner. The data set is passed through the decision tree and node Bayesian network for classification. The meta-learner (Functional tree classifier) will then select the value of the base classifier that has the higher accuracy based on majority voting. The key idea here is to always pick the value with higher accuracy since both base classifier (decision tree and Bayesian network) will always classify all instances. A performance evaluation was performed using a 10-fold cross validation technique on the individual base classifiers (decision tree and Bayesian network) and the ensemble classifier (DT-BN) using the KDD Cup 1999 dataset on WEKA tool. Experimental results show that the hybrid classifier (DT-BN) gives the best result in terms of accuracy and efficiency compared with the individual base classifiers (decision tree and BN). The decision tree gave a result of (99.9974% for DoS, 100% for Normal, 98.8069% for probing, 97.6021% for U2R and 73.0769% for R2L), the Bayesian network (99.6410% for DoS, 100% for Normal, 97.1756% for probing, 97.0693% for U2R and 69.2308% for R2L),while the ensemble method gave a result of (99.9977% for DoS, 100% for Normal, 98.8069% for probing, 97.6909% for U2R and 73.0769% for R2L).