Belawati Widjaja

Automatic lag selection in time series forecasting using multiple kernel learning

This paper reports the feasibility of employing the recent approach on kernel learning, namely the multiple kernel learning (MKL), for time series forecasting to automatically select the optimal lag length or size of sliding windows. MKL is an approach to choose suitable kernels from a given pool of kernels by exploring the combination of multiple kernels. In this paper, we extend the MKL capability to select the optimal size of sliding windows for time series domain by adopting the data integration approach which has been previously studied in the domain of image processing. In this study, each kernel represents the different lengths of time series lag. In addition, we also examine the feasibility of MKL for decomposed time series. We use the dataset from previous time series competitions as our benchmark. Our experimental results indicate that our approaches perform competitively compared to the previous methods using the same dataset. Furthermore, MKL may predict the detrended time series without explicitly computing the seasonality. The advantage of our method is in its ability in automatically selecting the optimal size of sliding windows and finding the pattern of time series.

Indistinguishable of AES-Based PRNG against Modification Attack Based on Statistical Distance Tests and Entropy Measures

Due to previous research, AES-based PRNG is not affected by insertion attack (in random manner) under level of significant α = 0.01, even it caused some failed tests in randomness. Completing the research, the writers do the observation of the modification attack in random manner against the output sequence of AES-based PRNG that is limited to 1-bit modification attack. The tests are performed by applying statistical distance test between the output sequence before and after the attack. To assure the attack effect, we also measure the entropy values of the sequence before and after attack and compare them. The attack scenario is still the same as the previous research [see 3], except replacing the insertion with modification and parameter  = 0.001. The results show that the modification attack does not give the significant effect on the randomness property of the AES-based PRNG. It was proved from 60 experiments of 1-bit modification attack, that the maximum statistical distances are still far away from = 0.001. And the change of the entropy source after the attack are very small and also still far away from the  = 0.001.

Gaussian Fuzzy Number for STR-DNA Similarity Calculation Involving Familial and Tribal Relationships

We performed locus similarity calculation by measuring fuzzy intersection between individual locus and reference locus and then performed CODIS STR-DNA similarity calculation. The fuzzy intersection calculation enables a more robust CODIS STR-DNA similarity calculation due to imprecision caused by noise produced by PCR machine. We also proposed shifted convoluted Gaussian fuzzy number (SCGFN) and Gaussian fuzzy number (GFN) to represent each locus value as improvement of triangular fuzzy number (TFN) as used in previous research. Compared to triangular fuzzy number (TFN), GFN is more realistic to represent uncertainty of locus information because the distribution is assumed to be Gaussian. Then, the original Gaussian fuzzy number (GFN) is convoluted with distribution of certain ethnic locus information to produce the new SCGFN which more represents ethnic information compared to original GFN. Experiments were done for the following cases: people with family relationships, people of the same tribe, and certain tribal populations. The statistical test with analysis of variance (ANOVA) shows the difference in similarity between SCGFN, GFN, and TFN with a significant level of 95%. The Tukey method in ANOVA shows that SCGFN yields a higher similarity which means being better than the GFN and TFN methods. The proposed method enables CODIS STR-DNA similarity calculation which is more robust to noise and performed better CODIS similarity calculation involving familial and tribal relationships.

Weighting for DNA Profiling

Identification of individual STR-based individuals is required for the investigation of Disaster Victim Identification and other applications. The DNA identification of an individual with the DNA of both biological parents, father, and mother, would result in a perfect match value, but what if the biological parents of the individual had died. In this research, we proposed a method of identifying DNA against an individual if one or both of the individual parents were absent, so it was necessary to match the individual DNA profiles with DNA profiles of existing family members. The conclusions from the results of individual DNA matching with DNA of family members were proposed using fuzzy inference system with weighted suggestion according to familial closeness.

Nonlinearly weighted multiple kernel learning for time series forecasting

Machine Learning methods such as Neural Network (NN) and Support Vector Regression (SVR) have been studied extensively for time series forecasting. Multiple Kernel Learning (MKL) which utilizes SVR as the predictor is yet another recent approaches to choose suitable kernels from a given pool of kernels by means of a linear combination of some base kernels. However, some literatures suggest that this linear combination of kernels cannot consistently outperform either the uniform combination of base kernels or simply the best single kernel. Hence, in this paper, other combination method is devised, namely the squared combination of base kernels, which gives more weight on suitable kernels and vice versa. We use time series data having various length, pattern and horizons, namely the 111 time series from NN3 competition, 3003 of M3 competition, 1001 of Ml competition and reduced 111 of Ml competition. Our experimental results indicate that our new forecasting approaches using squared combination of Multiple Kernel Learning (MKL) may perform well compared to the other methods on the same dataset.

CLS and CLS Close: The Scalable Method for Mining the Semi Structured Data Set

Semistructured pattern can be formally modeled as Graph Pattern. The most important problem to be solved in mining large semi structured dataset is the scalability of the method. With the successful development of efficient and scalable algorithms for mining frequent itemsets and sequences, it is natural to extend the scope of study to a more general pattern mining problem: mining frequent semistructured patterns or graph patterns. In this paper, we extend the methodology of pattern-growth and develop a novel algorithm called CLS (Canonical Labeling System), which discovers frequent connected subgraphs efficiently using either depth-first search or breadth-first search strategy.