Predicting seminal quality with artificial intelligence methods

After the publication of a meta-analysis directed by Elisabeth Carlsen (Carlsen, Giwercman, Keiding, & Skakkebaek, 1992), there is a debate about a possible decline in semen quality. Numerous studies show a decrease in semen parameters during the past two decades (Auger et al., 1995, Splingart et al., 2011, Swan et al., 1997, Swan et al., 2000), affecting the male fertility potential. Between the factors considered to explain this decline, there is an increase in the incidence of male reproductive diseases (Irvine, 2000, Splingart et al., 2011), but also has been suggested the effect of environmental or occupational factors (Giwercman and Giwercman, 2011, Wong et al., 2003), to a certain lifestyle (Agarwal et al., 2008, Martini, 2004).

To evaluate the male partner, clinicians use the data obtained from semen analysis (Kolettis, 2003), and they compare the obtained results with the corresponding reference value established by World Health Organization (WHO, 1999). Semen analysis is a good predictor of the male fertility potential (Bonde et al., 1998, Guzick et al., 2001, Slama et al., 2002, Zinaman et al., 2000), and is also necessary to evaluate candidates to become semen donors (Barratt et al., 1998, Carrell et al., 2002, Ecochard et al., 1999, Society British Andrology, 1999). There is a high variability in the testicular function of an individual (Keel, 2006), so it is recommended to interpret the results taking in account certain factors (i.e., fever, toxic exposure) that can modify the semen parameters (Rowe & Comhaire, 2000).

In this paper, assuming the influence of environmental factors and life habits in semen quality, we compare the prediction accuracy of three different artificial intelligence (AI) methods, Multilayer Perceptron (MLP), Support Vector Machines (SVM) and decision trees (DT), to determine the best Decision Support Systems (DSS) that can help in the evaluation of male fertility potential.

The current advances and progresses in the field of AI have led to the emergence of expert systems and DSSs for economics, linguistics, management science, mathematical modelling, psychology, etc.

There are good classifiers in the AI such as artificial neural network (ANN), DT (Polat & Gnes, 2009b), SVM (Conforti & Guido, 2010) or even hybrid methods that combine ANNs and fuzzy logic into fuzzy neural networks (FNN) (Kahramanli & Allahverd, 2008) which are widely used to aid medical diagnosis by means of decision support systems construction.

Moreover, the use of AI has also become widely accepted in medical applications. These methods include advantages as (a) Ease of optimisation, resulting in cost-effective and flexible non-linear modelling of large data sets; (b) Accuracy for predictive inference, with potential to support clinical decision making; (c) These models can make knowledge dissemination easier by providing explanation, for instance, using rule extraction or sensitivity analysis (Lisboa & Taktak, 2006).

The amount of information used in order to predict the male fertility potential makes really useful the application with AI methods not only for improving the accuracy but also to select the best features as, very often, the number of feature to deal with is huge (Gil and Johnsson, 2010a, Gil and Johnsson, 2010b, Gil and Johnsson, 2011, Gil et al., 2009, Gil et al., 2011, Polat and Gnes, 2009a, Subashini et al., 2009). This will lead to the knowledge discovery in databases, data mining or the process of extracting patterns from large data sets.

The main objective of this paper is to compare a number of classification methods applied to male fertility data sets. The approach chosen to address the problem is by the use of different AI methods, in particular DT, MLP and SVM. A comparative study of those will give us insight into the merits of the different methods when used on this problem.

The remaining part of the paper is organized as follows: First we start defining the materials and methods of the study (study population and the variables of the questionnaire). Then we continue with a brief description of the AI methods used in this paper: MLP, SVM and DT. Then we proceed by describing the design of our proposal and the experiments carried out in the result section; the available data as well as a detailed explanation of the different values of our database; Then we continue by describing the subsequent testing carried out in order to analyse the results; Finally we draw the relevant conclusions.