Data mining is defined as procedure of mining knowledge or hidden knowledge from past or historical data. 8. It is interactive process for discovering novel, valid, understandable and useful hidden pattern and relationship in data that may be used to make a valid prediction 7. The study used data mining classification algorithms which include the following:-

3.1 J48

J48 is an implementation of C4.5 in weka and it is one of the algorithms of data mining classification techniques. J48 build decision tree using the concept of information entropy from a set of training data in the same way as in id3. The training data is a set of S = s1, s2… sn of the ready-made classified samples. Each sample Si = x1, x2…. is a vector where x1, x2,…. xn represent features or attributes of the sample 9. Decision tree is a “divide-and-conquer” approach to the problem of learning from a set of independent instances, which leads naturally to a tree-like style of representation called a decision tree. 5, 6. At each node of the tree, J48 chooses one attribute of the data that most effectively splits its set of samples into subsets enriched in one class or the other. 9

3.2 Multilayer Perceptron

Multilayer Perceptron is one of the neutral network data mining classifier. It is well established Multilayer Perceptron as a promising alternative to various conventional classification techniques. Neutral networks are data driven self-adaptive techniques in they can adjust themselves to the data without explicit specification of distribution or fluxional form for underlying model 9.

3.3 BayesNet

BayesNet is usually called Bayesian network. It is a data mining predictive model that uses to reflect the states of the part of the world that is being modeled and describe how these states are related with probabilities. In order for a BayesNet to model a probability distribution, each variable is conditionally independent of all its no descendants in the graph given the value of all its parents. This implies

P(X1….. Xn) = ? P(Xi ?parents(Xi)) (1)

BayesNe is generally useful if it helps us to greater understand the world we are modeling, and it allows us to make useful predictions about how the world will behave 8.

3.4 Apriori algorithm

Apriori algorithm is association rule mining algorithm that used to find the frequent pattern, correlation, association and causal structure among items in a pool of data. Association rule is an implication of the form X ? Y, where X and Y are disjoint items i.e X ? Y = ?. The strength of association rule can be measured in items of its support and confidence. Support determines how often a rule is applicable to a given data set while confidence determine how frequently item in Y appear in transaction that X. 10

3.5 Evaluation of Classifiers

The data mining evaluation mechanism used for evaluating the performance of the various classification algorithms to identify the suitable algorithm to be applied in road traffic accident dataset for prediction include the following:-

Accuracy measures the proportion of the total number of predictions that were correct.

(2)

Specificity measures the proportion of actual negative cases which are correctly identified, as calculated using the equation:

(3)

Sensitivity measures the proportion of actual positive cases which are correctly identified, as calculated using the equation:

Precision measures the proportion of positive cases that were correctly identified, as calculated using the equation:

Recall measures the proportion of positive cases that were correctly identified, as calculated using the equation:

3.6 K-means Clustering

K-means clustering is used to group n objects into k-clusters such a way that the mean of the objects within the cluster are the nearest mean. The goal of this clustering method is to reduce the intra-cluster variance or the squared error. 15

3.7 Self-Organizing Map

Self-Organizing Map is unsupervised technique for visualizing high dimensional data with low dimension views. This dimensionality reduction method uses artificial neural network technique for discretized representation of the training data. 14

4. Experiment and Result

4.1 Data Set

The study used the data set of traffic road accident of Kano to Wudil high way in Nigeria. The data used covered the period of thirty months started from January, 2014 to June, 2016. The data set was used in the study of 7, 8 to predict the cause of accident, prone location and time along Kano to Wudil high way in Nigeria. The dataset contains four variables-Vehicle Type, Accident Time, Accident Cause and Accident Location.

Fig 1. Mosaic Display of the Accident Time against Accident Location

4.2 Experiment and Results

Weka data mining software was used for the experiment of the dataset of the study. Weka open source data mining software was used to mine the dataset. Weka contains machine learning algorithms for data mining tasks. The algorithm can either be called to java code or apply directly to a dataset. The study applied Multilayer Perceptron, J48, BayesNet classifiers or algorithms directly on 150 instances to traffic road accident data set. The result of the experiment is shown in Table 4.1.

Table 1. Comparison of different classifiers

Sl.

No.

Data mining Algorithm

(classifier )

Accuracy

Specificity

Sensitivity

Precision

Recall

1

Multilayer Perceptron

85.33%

0.473

0.853

0.848

0.853

2

BayesNet

80%

0.624

0.800

0.769

0.800

3

J48

78.67%

0.787

0.787

0.619

0.787

We had used the Apriori Rule Mining to find out the best possible association rules using Weka. We had found the following two rules and the result of the experiment is shown as below.

1. AccidentCause=WrongOvertaking AccidentLocation=LocationC 15 ==> VehicleType=SmallCar 15

2. AccidentTime=Evening AccidentCause=WrongOvertaking 32 ==> VehicleType=SmallCar 29

We had applied K-means clustering, Self Organizing Map (SOM) on the datasets as unsupervised learning using Orange data mining software. Orange is an open source data mining software for both novice and expert users with great visualization and large toolbox. The silhouette score of 0.7 was achieved to depict the meaningful clustering. The figures below visualized the clustering and unsupervised learning results.

Fig 2. SOM accident cause Fig 3. SOM accident location

Fig 4. Scatter Plot visualization of K-means clustering Fig 5. Silhouette Plot Clustered by ‘Accident Location’

5. Conclusion and Future Work

150 instances traffic road accident data set for Kano – Wudil high way road, in Nigeria using Weka was used to evaluate the performance of the three data mining algorithms. The algorithms were directly applied on the dataset. The results of the experiment of the study depicted that, for prediction on traffic road accident dataset, Multilayer Perceptron is most appropriate, suitable and efficient data mining algorithm to be used. In the course of the experiment, Multilayer Perceptron classifier performed well with 85.33% accuracy, followed by J48 with 78.66% accuracy and BayesNet had 80.66% accuracy. Therefore, study Multilayer Perceptron is recommended to scholars and researchers to be used as efficient data mining classifier or algorithm for predictive tasks. The study had also found two best rules for association rule mining using Apriori algorithm with 1.0 minimum supports and 1.27 minimum confidences for rule one and 0.91 minimum supports and 1.15 minimum confidences for rule two. K-means clustering and Self Organizing Map were also applied on the dataset with silhouette score of 0.7.The algorithms may be tested with more data and different datasets for the performance evaluation as a future work.

References

1 Dipo T. A. and Akinbola, O. (2012). Using Data Mining Technique to Predict Cause of Accident and Accident Prone Locations on Highways. American Journal of Database Theory and Application 2012, 1(3): 26-38

2 Elfadil A. M. (2014). Predicting Causes of Traffic Road Accidents Using Multi-class Support Vector Machines, Proceeding of the 10th International Conference on Data Mining, July 21-24, 2014, Las Vegas, Nevada,

USA, Page 37-42

3 Global status report on road safety, (2015). World Health Organization, Accessed from http://www.who.int/entity/violence_injury_prevention/publications/road_traffic/en/index.html Date: 25th December, 2017

4 Global status report on road safety: time for action, WHO, 2009.

5 Jaideep, K., and Chandra, P. S. (2016). Mining Road Traffic Accident Data to Improve Safety on Road-related Factors for Classification and Prediction of Accident Severity, International Research Journal of Engineering and Technology (IRJET), Vol. 03 (10)

6 Liling, L., Sharad, S. and Gongzhu, H. (2017). Analysis of road traffic fatal accidents using data mining techniques, Software Engineering Research, Management and Applications (SERA), 2017 IEEE 15th International Conference

7 Muhammad, L. J., Yakubu, A, Mohammed, I. A. (2017). Data Mining Driven Approach for Predicting Causes of Road Accident, 13th International Conference 2017- Information Technology for Sustainaible Development, Nigeria Computer Society, Vol. 28, pp. 10-15

8 Muhammad, L. J., Sani, S., Yakubu, A., Yusuf, M. M., Elrufai, T. A., Mohammed I. A. and Nuhu, A. M. (2017). Using Decision Tree Data Mining Algorithm to Predict Causes of Road Traffic Accidents, its Prone Locations and Time along Kano –Wudil Highway, International Journal of Database Theory and Application Vol.10, pp.197-206

9 Performance evaluation of the data mining models Accessed from www.shodhganga.inflibnet.ac.in/bitstream/10603/7989/14/14_chapter%205.pdf

Date: 29th December, 2017

10 Rakesh, A., Ramakrishnan, S., (1994). Fast Algorithms for Mining Association Rules in Large Databases, Proceedings of the 20th International Conference on Very Large Data Bases, p.487-499, September 12-15, 1994

11 Sachin, K. and Durga, T. (2016). A data mining approach to characterize road accident locations, Journal of Modern Transportation, Vol. 24 (1), pp 62–72

12 Scott D. and Andrew, M.(2008). Bayesian Networks: Independencies and Inference

13 Syeda, F. S., Mirza, M. A. and Baig, M.R. Pawar (2013).Performance Evaluation of Different Data Mining Classification Algorithm and Predictive Analysis, Journal of Computer Engineering (IOSR-JCE),Vol. 10(6), pp. 01-06

14 Ultsch, Alfred (2007). “Emergence in Self-Organizing Feature Maps”. In Ritter, H.; Haschke, R. Proceedings of the 6th International Workshop on Self-Organizing Maps (WSOM ’07). Bielefeld, Germany: Neuroinformatics Group. ISBN 978-3-00-022473-7.

15 Vattani., A. (2011). “k-means requires exponentially many iterations even in the plane”(PDF). Discrete and Computational Geometry. 45 (4): 596–616. doi:10.1007/s00454-011-9340-1.