Rafael Gomes Mantovani

To tune or not to tune: Recommending when to adjust SVM hyper-parameters via meta-learning

Many classification algorithms, such as Neural Networks and Support Vector Machines, have a range of hyper-parameters that may strongly affect the predictive performance of the models induced by them. Hence, it is recommended to define the values of these hyper-parameters using optimization techniques. While these techniques usually converge to a good set of values, they typically have a high computational cost, because many candidate sets of values are evaluated during the optimization process. It is often not clear whether this will result in parameter settings that are significantly better than the default settings. When training time is limited, it may help to know when these parameters should definitely be tuned. In this study, we use meta-learning to predict when optimization techniques are expected to lead to models whose predictive performance is better than those obtained by using default parameter settings. Hence, we can choose to employ optimization techniques only when they are expected to improve performance, thus reducing the overall computational cost. We evaluate these meta-learning techniques on more than one hundred data sets. The experimental results show that it is possible to accurately predict when optimization techniques should be used instead of default values suggested by some machine learning libraries.

Storage time prediction of pork by Computational Intelligence

We use of CI for storage time prediction only by pH, CIELab and WHC parameters.Our approach has pork assessment by a non-destructive, fast, accurate analysis.We evaluate the pork aging, accurately, without the analysis of lipid oxidation.We test prediction by J48, Naive Bayes, kNN, RF, SVM, MLP and Fuzzy. In this paper, a storage time prediction of pork using Computational Intelligence (CI) model was reported. We investigated a solution based on traditional pork assessment towards a low time-cost parameters acquisition and high accurate CI models by selection of appropriate parameters. The models investigated were built by J48, Naive Bayes (NB), k-NN, Random Forest (RF), SVM, MLP and Fuzzy approaches. CI input were traditional quality parameters, including pH, water holding capacity (WHC), color and lipid oxidation extracted from 250 samples of 0, 7 and 14days of post mortem. Five parameters (pH, WHC, Lź, aź and bź) were found superior results to determine the storage time and corroborate with identification in minutes. Results showed RF (94.41%), 3-NN (93.57%), Fuzzy Chi (93.23%), Fuzzy W (92.35%), MLP (88.35%), J48 (83.64%), SVM (82.03%) and NB (78.26%) were modeled by the five parameters. One important observation is about the ease of 0-day identification, followed by 14-day and 7-day independently of CI approach. Result of this paper offers the potential of CI for implementation in real scenarios, inclusive for fraud detection and pork quality assessment based on a non-destructive, fast, accurate analysis of the storage time.

Effectiveness of Random Search in SVM hyper-parameter tuning

Classification is one of the most common machine learning tasks. SVMs have been frequently applied to this task. In general, the values chosen for the hyper-parameters of SVMs affect the performance of their induced predictive models. Several studies use optimization techniques to find a set of hyper-parameter values that induces classifiers with good predictive performance. This paper investigates the hypothesis that a simple Random Search method is sufficient to adjust the hyper-parameters of SVMs. A set of experiments compared the performance of five tuning techniques: three meta-heuristics commonly used, Random Search and Grid Search. The experimental results show that the predictive performance of models using Random Search is equivalent to those obtained using meta-heuristics and Grid Search, but with a lower computational cost.

Hyper-Parameter Tuning of a Decision Tree Induction Algorithm

Supervised classification is the most studied task in Machine Learning. Among the many algorithms used in such task, Decision Tree algorithms are a popular choice, since they are robust and efficient to construct. Moreover, they have the advantage of producing comprehensible models and satisfactory accuracy levels in several application domains. Like most of the Machine Leaning methods, these algorithms have some hyper-parameters whose values directly affect the performance of the induced models. Due to the high number of possibilities for these hyper-parameter values, several studies use optimization techniques to find a good set of solutions in order to produce classifiers with good predictive performance. This study investigates how sensitive decision trees are to a hyper-parameter optimization process. Four different tuning techniques were explored to adjust J48 Decision Tree algorithm hyper-parameters. In total, experiments using 102 heterogeneous datasets analyzed the tuning effect on the induced models. The experimental results show that even presenting a low average improvement over all datasets, in most of the cases the improvement is statistically significant.

Pattern Recognition of Lower Member Skin Ulcers in Medical Images with Machine Learning Algorithms

Misleading diagnosis of skin diseases may result in complications during the healing process. Skin images provide an important contribution to medical staff on storing and exchanging information to try preventing misdiagnosis. For such, image segmentation process may benefit from use of machine learning techniques, increasing simplicity of procedure, reducing computational costs and improving the diagnosis. This paper presents a comparison among different paradigms of machine learning to validate the segmentation of medical images of lower members ulcers, this segmentation allows wound pattern recognition to determinate injury region aiming at reducing the subjectivity of human evaluation.

A Meta-Learning Approach for Recommendation of Image Segmentation Algorithms

There are many algorithms for image segmentation, but there is no optimal algorithm for all kind of image applications. To recommend an adequate algorithm for segmentation is a challenging task that requires knowledge about the problem and algorithms. Meta-learning has recently emerged from machine learning research field to solve the algorithm selection problem. This paper applies meta-learning to recommend segmentation algorithms based on meta-knowledge. We performed experiments in four different meta-databases representing various real world problems, recommending when three different segmentation techniques are adequate or not. A set of 44 features based on color, frequency domain, histogram, texture, contrast and image quality were extracted from images, obtaining enough discriminative power for the recommending task in different segmentation scenarios. Results show that Random Forest meta-models were able to recommend segmentation algorithms with high predictive performance.

Machine Learning Applied to Near-Infrared Spectra for Chicken Meat Classification

Identification of chicken quality parameters is often inconsistent, time-consuming, and laborious. Near-infrared (NIR) spectroscopy has been used as a powerful tool for food quality assessment. However, the near-infrared (NIR) spectra comprise a large number of redundant information. Determining wavelengths relevance and selecting subsets for classification and prediction models are mandatory for the development of multispectral systems. A combination of both attribute and wavelength selection for NIR spectral information of chicken meat samples was investigated. Decision Trees and Decision Table predictors exploit these optimal wavelengths for classification tasks according to different quality grades of poultry meat. The proposed methodology was conducted with a support vector machine algorithm (SVM) to compare the precision of the proposed model. Experiments were performed on NIR spectral information (1050 wavelengths), colour (CIE , chroma, and hue), water holding capacity (WHC), and pH of each sample analyzed. Results show that the best method was the REPTree based on 12 wavelengths, allowing for classification of poultry samples according to quality grades with precision. The selected wavelengths could lead to potential simple multispectral acquisition devices.

Applying multi-label techniques in emotion identification of short texts

Abstract Sentiment Analysis is an emerging research field traditionally applied to classify opinions, sentiments and emotions towards polarity and subjectivity expressed in text. An important characteristic to automatic emotion analysis is the standpoint, in which we can look at an opinion from two perspectives, the opinion holder (author) who express an opinion, and the reader who reads and perceives the opinion. From the reader’s standpoint, the interpretations of the text can be multiple and depend on the personal background. The multiple standpoints cognition, in which readers can look at the same sentence, is an interesting scenario to use the multi-label classification paradigm in the Sentiment Analysis domain. This methodology is able to handle different target sentiments simultaneously in the same text, by also taking advantage of the relations between them. We applied different approaches such as algorithm adaptation, problem transformation and ensemble methods in order to explore the wide range of multi-label solutions. The experiments were conducted on 10,080 news sentences from two different real datasets. Experimental results showed that the Ensemble Classifier Chain overcame the other algorithms, average F-measure of 64.89% using emotion strength features, when considering six emotions and neutral sentiment.

Incorporating instance correlations in multi-label classification via label-space

Multi-label classification is a machine learning task where instances can be classified into two or more labels simultaneously. In this task, there exist correlations between the instances belonging to same or similar sets of labels. This paper proposes the incorporation of instance correlations by modifying the multi-label datasets. We used the label-space to create new features, which represent these correlations. The original and modified datasets were used with different multi-label classification methods. Experiments have shown that better results can be obtained when instance correlations were incorporated in the classification tasks. All methods were evaluated with measures specifically designed for multi-label problems.

Decision Trees for the Detection of Skin Lesion Patterns in Lower Limbs Ulcers

Misleading diagnosis of skin diseases can result in complications during the healing process. Skin images provide important information for the medical staff for information storage and exchange, to trying to prevent this misdiagnosis from happening. For such, a good segmentation process is needed. The segmentation of these images is already being used and has been an effective tool for skin diseases recognition. This paper presents a method for targeting seeds for region growing algorithms, as several of region growing algorithms have good clustering results, but are sensitive to seed. Machine learning were use to create the seed for segmentation of medical images of skin ulcers in the lower limbs. For machine learning, decision tree algorithms were used, which bring a more intuitive approach. The results were compared with gold standard obtained with the help of experts, the results were good and opened paths that can be followed for further work since, even though good results, they can still be improved.