AI News, BOOK REVIEW: Ordinal Prediction using machine learning methodologies ... artificial intelligence

Using machine learning methods to determine a typology of patients with HIV-HCV infection to be treated with antivirals

In this way, a 5-fold cross-validation method optimising the AUC measure, was run to select the best value for the penalty of the error (C) and for the RBF kernel coefficient (γ), both chosen within the range {10−4, 10−3, …102}.

The results obtained were 0.755 in terms of CCR and 0.716 for AUC, being clearly worse than the ones obtained by the models in Table 2, whereas in terms of MS, it led to 0.603, slightly better than the MS obtained by the ANN models.

Attention is drawn to the importance of input variable X4 (“recent PWIDs”) among the classifiers used, since it was included in all the best models and its non-inclusion in the RBFNN model (called RBFNN2) reduced the mean AUC from 0.795 ± 0.005 to 0.483 ± 0.028 and the mean MS from 0.550 ± 0.008 to 0.014 ± 0.019, making it a trivial classifier that classifies most instances in one class.

It is also worth noting that the use of just six input variables makes the model easy to interpret, easy to implement and requires little training time, while the rest of the techniques need more than 6 input variables.

The ROC curve provides a graphical display of true positives (TPR) and false positives (FPR) on the x− and y− axes, respectively, where TPR is equivalent to sensitivity, and FPR is equal to 1 − specificity for varying cut-off points of test probability values.

However, it would make the models deal with two additional disadvantages: a important lost of interpretability, since the new input variables are combinations of the original ones, and a possible lost of performance, due to the need of more robust and accurate information.

The average results of the 30 runs are 0.752 ± 0.007 in terms of CCR, 0.550 ± 0.026 in terms of MS and 0.767 ± 0.008 in terms of AUC, which are worse than the results obtained by all the models with the original datasets.

To consider the statistical significance of differences between means (CCR, MS, AUC and #conn) for each ANN topology (SUNN, PUNN and RBFNN), the non-parametric Kolmogorov-Smirnov (K-S) test for normality was used with α = 0.05, to evaluate whether CCR, MS, AUC and #conn followed a normal distribution.

Dynamically weighted evolutionary ordinal neural network for solving an imbalanced liver transplantation problem

HighlightsThe problem of constructing an organ allocation decision-support system combining donor, recipiend and surgery characteristics using artificial neural networks is assessed.A dynamically weighted evolutionary algorithm alleviates the imbalanced nature of the dataset.Ordinal over-sampling techniques help balancing the training set and improve the results obtained by the classifiers.An ordinal artificial neural network has been proven to perform well for the 4-category classification problem assessed.An extended supranational experimental design for liver transplantation allocation could be feasible considering more countries.

To solve this, we combine two approaches, a cost-sensitive evolutionary ordinal artificial neural network (ANN) (in which we propose to incorporate dynamic weights to make more emphasis on the worst classified classes) and an ordinal over-sampling technique (which adds virtual patterns to the minority classes and thus alleviates the imbalanced nature of the dataset).

ConclusionsThe combination of the proposed cost-sensitive evolutionary algorithm together with the application of an over-sampling technique improves the predictive capability of our model in a significant way (especially for minority classes), which can help the surgeons make more informed decisions about the most appropriate recipient for an specific donor organ, in order to maximize the probability of survival after the transplantation and therefore the fairness principle.

Constructing and Combining Orthogonal Projection Vectors for Ordinal Regression

Ordinal regression is to predict categories of ordinal scale and it has wide applications in many domains where the human evaluation plays a major role.

In this paper, we propose a novel ordinal regression strategy which consists of two stages: firstly orthogonal feature vectors are extracted and then these projector vectors are combined to learn an ordinal regression rule.