Dropping out of school is a problem that affects students around the world, with consequences for society. Therefore, in this study we will try to analyze this situation in higher education that affects not only students, but also different institutions training courses. By applying a pattern recognition system, feature selection methods are implemented to find those factors that could significantly influence the school career of university students at higher education institution in Portugal. These methods correspond to the so-called Features Selection Methods and Wrapper Methods, which allow analyzing sets with large volumes of data and which in turn correspond to mixed data types (numeric, categorical, object, etc.). The complexity of each of these methods used mostly involves all the features of the set, which will allow more accurate results to be obtained. After analyzing the results in the six methods used for Feature Selection, we can observe that in the Fisher Index, Pearson Correlation and Forward Selection, the characteristics of previous grades are more relevant to determine the success or failure of the student. For its part, in Variance Selection, Backward Selection and Exhaustive Selection, the characteristics that had the greatest weight were those related to the background and previous courses of the father and mother. With the training results, the resulting features with the Nayve Bayes and Nayve Bayes with Neyman Pearson models have an accuracy rate of around 70%. The case of the Hidden Markov model with its accuracy rate of 32.5% does not imply a Gaussian criterion, which gives us a lower accuracy, but no less reliable.