Ruji P. Medina

A hybrid approach towards improved artificial neural network training for short-term load forecasting

The power of artificial neural networks to form predictive models for phenomenon that exhibit non-linear relationships is a given fact. Despite this advantage, artificial neural networks are known to suffer drawbacks such as long training times and computational intensity. The researchers propose a two-tiered approach to enhance the learning performance of artificial neural networks for phenomenon with time series where data exhibits predictable changes that occur every calendar year. This paper focuses on the initial results of the first phase of the proposed algorithm which incorporates clustering and classification prior to application of the backpropagation algorithm. The 2016--2017 zonal load data of France is used as the data set. K-means is chosen as the clustering algorithm and a comparison is made between Naïve Bayes and k-Nearest Neighbors to determine the better classifier for this data set. The initial results show that electrical load behavior is not necessarily reflective of calendar clustering even without using the min-max temperature recorded during the inclusive months. Simulating the day-type classification process using one cluster, initial results show that the k-nearest neighbors outperforms the Naïve Bayes classifier for this data set and that the best feature to be used for classification into day type is the daily min-max load. These classified load data is expected to reduce training time and improve the overall performance of short-term load demand predictive models in a future paper.

Towards academic affect modeling through experimental hybrid gesture recognition algorithm

The identification of learner engagement is an important aspect of assessment. Aside from facial expressions, gesture is a key feature in the identification of student engagement. The costly video invigilation during assessment shows the need to find other ways to define student engagement during an online examination. For this purpose, this study proposed gesture modeling to classify and identify affect. The research defines student disengagement affect using head poses as gesture during the online examination. The divide-and-conquer algorithm implementation on object detection using Haar Cascade feature extraction and HMM classification resulted in 78.77% accuracy level to classify disengaged behavior during an online examination. The experimental results show that head-poses when properly modeled can be used to define affect as applied to examination behavior.

Enhanced RC6 permutation-diffusion operation for image encryption

RC6 is considered one of the encryption strategies being utilized and modified to be suited for image protection. However, the RC6 encryption algorithm is still prone to statistical and differential attack which is one of the major concerns whenever open networks are to be used. In this paper, we proposed a modification of RC6 algorithm for image encryption. A new technique of permutation-diffusion architecture based on a modified concept of cyclic shift is integrated to increase the degree of permutation and diffusion mechanism of RC6 for image encryption. The security level, processing time and diffusion rate were tested using image evaluation metrics such as correlation coefficient test, number of pixel change rate (NPCR) test, unified average change intensity (UACI) test, and run-time analysis to measure its efficiency along with the simulation process. Results show that the proposed method illustrates a lower value of correlation among pixels and higher value of NPCR and UACI which makes the proposed method highly resistant to statistical and differential attacks. Increased security level with minimal time consumption is achieved in our approach.

Application of enhanced expectation maximization (EnEM) algorithm for image segmentation

An Enhanced EM (EnEM) algorithm was developed through the integration of the concept of firefly movement and light intensity of Firefly Algorithm in its initialization stage. The improved initial parameter selection technique of EnEM algorithm leads to a better clustering performance when applied to image segmentation. The procedure converts first the image from RGB to HSV color space. A saturation threshold function was used in labelling the pixel and performed median filter for post processing to eliminate the noisy pixel to produce the final segmented image. An image segmentation module was developed and different test images were used. Experiments show that the application of EnEM to image segmentation produces lower MSE and higher PSNR which leads to a better segmentation.

ICD-9 Tagging of Clinical Notes Using Topical Word Embedding

Medical records, which contains text, has been dramatically increasing everyday. This means that there is a greater need of analyzing health information in a better way. And this can be done through document classification in natural language applications. In this study, we describe tagging of patient notes with ICD-9 codes through topical word embedding in deep learning called EnHANs. We formulate this paper as a multi-label, multi-class classification problem to categorize the ICD-9 codes of a dataset with 400,000 critical care unit medical records. Knowing accurate diagnosis using ICD-9 codes is a vital information for billing and insurance claims. We demonstrate that through the use of topical word embedding model, we learn to classify patient notes with their corresponding ICD-9 labels moderately well than single-label classification.

An Adequate Dietary Planning Model Using Particle Swarm Optimization

This study aims to develop a linear programming optimization model that will effectively assist dietitians in preparing a meal plan for adults with the variety of foods that include appropriate food group proportion and at the same time meets his/her total daily energy requirement, macronutrients and micronutrients needs. The objective function of the programming model is designed to minimize food cost. The model was solved by Particle Swarm Optimization written in Matlab. As a result, a low-cost meal for a day was selected.

Enhanced Manhattan-Based Clustering Using Fuzzy C-Means Algorithm

Fuzzy C-Means is a clustering algorithm known to suffer from slow processing time. One factor affecting this algorithm is on the selection of appropriate distance measure. While this drawback was addressed with the use of the Manhattan distance measure, this sacrifice its accuracy over processing time. In this study, a new approach to distance measurement is explored to answer both the speed and accuracy issues of Fuzzy C-Means incorporating trigonometric functions to Manhattan distance calculation. Upon application of the new approach for clustering of the Iris dataset, processing time was reduced by three iterations over the use of Euclidean distance. Improvement in accuracy was also observed with 50% and 78% improvement over the use of Euclidean and Manhattan distances respectively. The results provide clear proof that the new distance measurement approach was able to address both the slow processing time and accuracy problems associated with Fuzzy C-Means clustering algorithm.

Subset Sum-Based Verifiable Secret Sharing Scheme for Secure Multiparty Computation

Despite the information theoretic security of Shamir Secret Sharing Scheme and the ideality of Verifiable Secret Sharing Scheme in ensuring the honesty of a dealer of the shared secret and the shared secret itself, the detection and removal of an adversary posing as shareholder is still an open problem due to the fact that most of the studies are computationally and communicationally complex. This paper proposes a verifiable secret sharing scheme using a simple subset sum theory in monitoring and removing compromised shareholder in a secure multiparty computation. An analysis shows that the scheme cost minimal computational complexity of O(n) on the worst-case scenario and a variable-length communication cost depending on the length of the subset and the value of n.

Towards enhanced hierarchical attention networks in ICD-9 tagging of clinical notes

Text is an important element in document classification in many natural language applications. Natural language processing (NLP) is todays computational advancement that provides many significant modern uses of text documents such as efficient information retrieval. In this paper, we describe the theoretical framework of predicting ICD-9 codes through tagging of clinical notes using our improved framework in deep learning called EnHANs. This proposed model improvement covers combination of word and topic embedding, as well as adding character-level representation of a document in a hierarchical attention neural networks. This paper also present the use of sigmoid activation function in the last layer of the enhanced neural network in order to arrive with a multi-label, multi-class prediction of clinical notes with ICD-9 codes.

A Unique One-Time Password Table Sequence Pattern Authentication: Application to Bicol University Union of Federated Faculty Association, Inc. (BUUFFAI) eVoting System

Electronic Voting System (EVS) is a type of voting program that deals primarily with the selection, the casting of votes with embedded security mechanism that detects errors, and the tamper-proof election of results done through the use of an electronic system. It can include optical scan, specialized voting kiosks and Internet voting approach. Most organizations have difficulties when it comes to voting and the Bicol University Union of Federated Faculty Association Incorporated (BUUFFAI) is not an exception. Some of the problems involved include convenience, cost, geographical location of the polling precinct, and voting turnouts. This study extends the scope of the current BUUFFAI eVoting system to address such issues and to eliminate inconvenience both to the faculty voters and the facilitators. This voting scheme used an algorithmic OTP scheme based on table sequence pattern schedule that randomly generates an XY coordinate unique to voters that will be sent to voter registered email address. This study addressed the security requirements and maintained election procedures with confidentiality, integrity and availability.