Search

Purpose: This study aims to differentiate the quality of salak fruit with machine learning. Salak is classified into two classes, good and bad class.Design/methodology/approach: The algorithm used in this research is transfer learning with the VGG16 architecture. Data set used in this research consist of 370 images of salak, 190 from good class and 180 from bad class. The image is preprocessed by resizing and normalizing pixel value in the image. Preprocessed images is split into 80% training data and 20% testing data. Training data is trained by using pretrained VGG16 model. The parameters that are changed during the training are epoch, momentum, and learning rate. The resulting model is then used for testing. The accuracy, precision and recall is monitored to determine the best model to classify the images.Findings/result: The highest accuracy obtained from this study is 95.83%. This accuracy is obtained by using a learning rate = 0.0001 and momentum 0.9. The precision and recall for this model is 97.2 and 94.6.Originality/value/state of the art: The use of transfer learning to classify salak which never been used before.

Solid waste problem become a serious issue for the countries around the world since the amount of generated solid waste increase annually. As an effort to reduce and reuse of solid waste, a classification of solid waste image is needed to support automatic waste sorting. In the image classification task, image segmentation and feature extraction play important roles. This research applies recent deep leaning-based segmentation, namely pyramid scene parsing network (PSPNet). We also use various combination of image feature extraction (color, texture, and shape) to search for the best combination of features. As a comparison, we also perform experiment without using segmentation to see the effect of PSPNet. Then, support vector machine (SVM) is applied in the end as classification algorithm. Based on the result of experiment, it can be concluded that generally applying segmentation provide better source for feature extraction, especially in color and shape feature, hence increase the accuracy of classifier. It is also observed that the most important feature in this problem is color feature. However, the accuracy of classifier increase if additional features are introduced. The highest accuracy of 76.49% is achieved when PSPNet segmentation is applied and all combination of features are used.

Deep learning has shown promising results in various text-based classification tasks. However, deep learning performance is affected by the number of data, i.e., when the number of data is small, deep learning algorithms do not perform well, and vice versa. Classical machine learning algorithms commonly work well for a few data, and their performance reaches an optimal value and does not increase with the increase in sample data. Therefore, this study aimed to compare the performance of classical machine learning and deep learning methods to detect temperament based on Indonesian Twitter. In this study, the proposed Indonesian Linguistic Inquiry and Word Count were employed to analyze the context of Twitter. The classical machine learning methods implemented were support vector machine and K-nearest neighbor, whereas the deep learning method employed was a convolutional neural network (CNN) with three different architectures. Both learning methods were implemented using multiclass classification and one versus all (OVA) multiclass classification. The highest average f-measure was 58.73%, obtained by CNN OVA with a pool size of 3, a dropout value of 0.7, and a learning rate value of 0.0007.

Solid waste management issue is main problem especially in developing countries, including Indonesia. Several efforts are made to solve waste management problem. Indonesia government has launched movement to sort different type of garbage on September 2019. Automatic garbage sortation is able to help this program. In order to be able to perform this task, the computer needs to differentiate each type of garbage. This process can be done by using machine learning method to differentiate garbage type. In this research, Transfer learning is used to perform classification task on TrashNet dataset. The models used in this research are ImageNet pretrained VGG16, ResNet-50 and Xception.The experiment result shows that Xception model is able to achieve highest accuracy of 88%, average precision of 84%, and average recall of 84%

Snake fruit, or most famous as Salak, is Indonesian local fruit. Salak is also one of fruit commodity from Indonesia. To perform export on Salak, rigid sortation is performed. The sortation is usually done manually. This study will implement digital image processing technique to differentiate Salak quality for export purpose. Salak sample were taken from Magelang district, one of the largest Salak producer. The feature used in this study is Histogram of Oriented Gradient. The classification used is Extreme Learning Machine (ELM). It is shown in this study that by using ELM, the highest accuracy can be achieved is 95%. A comparison classifier, SVM, is also used in this study. In this case SVM is able to achieve highest accuracy of 97.3%, which is still higher than ELM result

Microarray data classification has a great challenge due to number of samples which is much smaller compared to the number of genes. The problem is getting harder when the dataset has multiclass target and the number of samples in each class is not well distributed (which is called imbalance data distribution). In this research, two different approaches to handle imbalance data distribution are studied, they are SMOTE (based on data approach) and weighted ELM (based on algorithmic approach). To evaluate the performance of the proposed method, two public imbalanced multiclass microarray dataset are used, GCM (Global Cancer Map) and Subtypes-Leukemia dataset. The results of experiment show that the implementation of SMOTE and weighted ELM on GCM dataset have no significant effect in the classification performance. Different with the Subtypes-Leukemia dataset, the implementation of SMOTE and weighted ELM has improved the classification performance compared to the previous research. Generally, the results show that weighted ELM perform slightly better compared to SMOTE to increase the accuracy of the minority class.

The recognition of handwritten characters remain one of challenging task in character recognition problems. The variations created by each person in writing the characters affect the character recognition result. Many studies have been performed to increase the performance of Javanese character recognition. The efforts are to extract the best feature for classification or to get the best classifier for classification. In this study, HOG feature and Zoning Based Feature will be used to classify Javanese Characters. The performance of both features will be compared for classifying Javanese character by using SVM classifier. Two types of inputs will be used for each feature extractor, binary and skeleton of the character image. The experiment showed that HOG feature is able to show higher accuracy as compared to the simple zone based feature (88.45%). The best accuracy for HOG is achieved by using binary input. On the other hand, despite its simplicity zone based feature is able to achive 81.98% accuracy by using skeleton input. Considering that the zone based feature used in this research is simply the pixel count in each zone, future research may be performed to extract more statistical properties on each zone. Future works may also focus on rotation free feature extraction for Javanese character classification.

Javanese character is traditional character from Indonesia. The character is used to write most of literature works from Kingdom era in Indonesia, for example Majapahit Kingdom and Mataram Kingdom. However, due to its rarely use in daily activities, most Javanese are now unable to recognize Javanese characters. The importance of Javanese characters in Indonesia's culture motivates some researchers to perform study on this topic. Several researches have been performed to classify Javanese characters with various method, however the accuracy are still low. On the other hand, deep learning has performed well in image classification. In this paper, the use of deep learning technique to classify handwritten Javanese characters is studied. The classification used 2470 images from 20 javanese characters. The size of input image used is 32×32 pixel. The classification is performed by using convolutional neural network (CNN) and Deep Neural Network (DNN). The accuracy with k-fold cross validation obtained are 70.22% and 64.65% for CNN and DNN, respectively.

Microarray data are now often used as an alternative in the classification of cancer classes. The challenges in microarray data classification are the huge number of genes and limited samples, so dimensionality reduction and classification algorithm play important roles in building the model. In this research, ReliefF feature selection is applied to reduce the dimensionality of microarray data, then extreme learning machine (ELM) was applied as a classification method. Two benchmark multiclass microarray dataset, GCM and Sup types-Leukemia, are used in this research to evaluate the proposed method. In order to reduce the bias in the end result, 5-cross validation was applied only on training data to select the best combination of parameter values. Then, to evaluate the performance of the proposed method, the experiment of training and testing is repeated ten times using the randomly split of training and testing data and using only the best combination of parameter values. Then, the proposed method is evaluated in term of accuracy and sensitivity. The proposed method show the improvement in accuracy, compare to the previous research, both on GCM and Subtypes-Leukemia dataset. But, the sensitivity among all classes are still not well averaged. The sensitivity become worse on the class of minority sample.