IMPROVE MACHINE LEARNING RESULTS FOR SEMEN ANALYSIS USING ENSEMBLE META CLASSIFICATION

: Numerous studies have found that, declining male fertility around the world in the past 50 years are still falling at a rate of two per cent every year. Reasons for this decline range from our increasingly stressful lifestyles, poor diet, drinking and smoking, or a lack of exercise and environmental factors. Applications of machine learning techniques have been implemented in many fields, including health care. This paper have experimented ensemble meta classification techniques such as bagging, boosting and stacking for improving the performance of the Machine Learning algorithms such as Decision-tree (J48), IBK and Naïve-Bayesian (NB) classification for the characterization of seminal quality. The performance of general machine learning classifier is compared with ensemble classifier models and also been verified with their accuracy and error rate.


I. INTRODUCTION
Over recent years there has been an increasing concern that a decline in male fertility is occurring. There have been many suggestions for the decline in developed countries. Climatic conditions, less exercise, excess weight gain, poor food choices, alcohol and tobacco use; these are just some of the environmental and life style factors that are proven to adversely impact fertility [1] [2]. Hence the researchers carried out the semen analysis which is important for the assessment of male fertility potential and it can also be used for the assessment of sperm donors.
Machine learning has been incorporated in various areas including help with medical tasks such as disease identification and diagnosis, personalized treatment and behavioral modification [3] [4]. This is run mainly by supervised learning that allows physicians to decide from limited sets of diagnoses or estimate patient risks based on symptoms and genetic information [5]. Machine learning application also helps pharmaceutical companies in their pursuit to find better ways of discovering drugs and manufacturing them [6][7] [8].
The main objective of this paper is to improve the machine learning results using ensemble Meta classifier model, as applied to the problem of semen quality categorization. Here male fertility data set has been taken from the UCI data set repository. The rest of the paper is organized as follows. In section 2 data set have been discussed, in section 3 the methodology for ensemble meta classification have been discussed with experiment's design, section 4 shows the analysis of the results. Finally, section 5 contains conclusion with the direction of future work.

II. DATA SET
The fertility data set is taken from the University of California Irvine (UCI) dataset repository. It consists of semen samples, obtained from 100 volunteers, and analyzed according to WHO 2010 criteria. The data set attributes are based on the fact that sperm concentration is affected by the social demographic and environmental factors, health status and life style habits. The data set can be summarized as follows: • Number of Attributes: 9 plus the class attribute • Number of instances: 100 • Missing attribute values: None • Class distribution: There are 88 normal samples (88%) and 12 altered samples (12%) Table I

b) Boosting
Boosting is a sequential technique in which, the first algorithm is trained on the entire dataset and the subsequent algorithms are built by fitting the residuals of the first algorithm, thus giving higher weight to those observations that were poorly predicted by the previous model.

c) Stacking
In stacking multiple layers of machine learning models are placed one over another where each of the models passes their predictions to the model in the layer above it and the top layer model takes decisions based on the outputs of the models in layers below it. Two layers of machine learning models such as : i) Bottom layer models (d 1 , d 2 , d 3 ) which receive the original input features(x) from the dataset. ii) Top layer model, which takes the output of the bottom layer models (d 1 , d 2 , d 3 ) as its input and predicts the final output.

B. Model Selection
In this paper, three models are selected such as Decision Tree, Naïve-Bayesian and IBK (K-Nearest Neighbor) for classification of fertility data. These are the different types of classification techniques work differently for different datasets. Some techniques give better efficiency for a dataset of very large size but it might not be the optimal technique to use for a dataset with higher number of attributes. Each model is a weak learner which might not be good for the entire dataset but is good for some part of the dataset. Thus, each model actually boosts the performance and accuracy by the ensemble meta classifiers. Hence, Table  II shows the accuracy metrics for each classifier.

IV. RESULTS AND DISCUSSION
After improving the accuracy of the classifiers, the best model has been determined from 10-fold cross-validation. 10-fold cross validation is a measure to evaluate the accuracy of the classifiers or predictors in terms of error. It may seem intuitive to select the model with the lowest error rate; however, the mean error rates are just estimates of error on the true population of fertility data cases. Although the mean error rates obtained for three models may appear different, that difference may not be statistically significant.
The true positives, true negatives, false positives, and false negatives are also useful in assessing the costs and benefits (or risks and gains) associated with a classification model. The cost associated with a false negative (such as, incorrectly predicting that an Altered person is Normal) is far greater than that of a false positive (incorrectly yet conservatively labeling a Normal person as Altered Person). Loss functions measure the error between y i and the predicted value, y i '.