Candidate Generation Process Research Articles

Mining Weighted Patterns from Time Series Databases Based on Sliding Window

Data mining has traditionally relied heavily on sliding window-based challenges, which has sparked a variety of studies. For each new window in time series mining, current literature mandates the rebuilding of the underlying structure, Suffix Tree - A trie-based structure representing all the suffixes of a string. However, reconstruction struggles when the window is wide or when sliding happens frequently. As a result, we provide a new technique Dynamic Tree-Based Approach to handle Sliding Windows (DTSW) in time series in this study that dynamically changes the representative suffix tree structure rather than reconstructing it after every alteration or sliding. In addition, we also put forth a different approach to the issue of extracting weighted periodic patterns from time series. To prevent testing pointless patterns, existing studies mostly rely on the weight of the database's highest-weighted item. However, these methods continue to examine numerous patterns. These methods still examine numerous patterns to see whether they can be candidates. Our proposed measure Maximum Possible Weighted Support (MPWS) accelerates the candidate generation process by removing numerous unnecessary patterns in advance. The novelty of MPWS is it considers the maximum weighted average over the maximum weighted item extension by enforcing more constraints. The usefulness of our two techniques in handling sliding windows and trimming redundant candidate patterns is demonstrated by experimental results using a variety of real-world datasets. Our experiments state that our dynamic handling technique significantly improves runtime than the reconstruction in a dynamic sliding window-based environment with simultaneous insertion and deletion actions and MPWS reduces the number of tested patterns resulting in lesser mining time in weighted time series pattern mining. DUJASE Vol. 8 (1) 13-25, 2023 (January)

De Novo Drug Design Using Transformer-Based Machine Translation and Reinforcement Learning of an Adaptive Monte Carlo Tree Search.

The discovery of novel therapeutic compounds through de novo drug design represents a critical challenge in the field of pharmaceutical research. Traditional drug discovery approaches are often resource intensive and time consuming, leading researchers to explore innovative methods that harness the power of deep learning and reinforcement learning techniques. Here, we introduce a novel drug design approach called drugAI that leverages the Encoder-Decoder Transformer architecture in tandem with Reinforcement Learning via a Monte Carlo Tree Search (RL-MCTS) to expedite the process of drug discovery while ensuring the production of valid small molecules with drug-like characteristics and strong binding affinities towards their targets. We successfully integrated the Encoder-Decoder Transformer architecture, which generates molecular structures (drugs) from scratch with the RL-MCTS, serving as a reinforcement learning framework. The RL-MCTS combines the exploitation and exploration capabilities of a Monte Carlo Tree Search with the machine translation of a transformer-based Encoder-Decoder model. This dynamic approach allows the model to iteratively refine its drug candidate generation process, ensuring that the generated molecules adhere to essential physicochemical and biological constraints and effectively bind to their targets. The results from drugAI showcase the effectiveness of the proposed approach across various benchmark datasets, demonstrating a significant improvement in both the validity and drug-likeness of the generated compounds, compared to two existing benchmark methods. Moreover, drugAI ensures that the generated molecules exhibit strong binding affinities to their respective targets. In summary, this research highlights the real-world applications of drugAI in drug discovery pipelines, potentially accelerating the identification of promising drug candidates for a wide range of diseases.

Efficient mining of high utility itemsets with multiple minimum utility thresholds

Mining high utility itemsets is considered to be one of the important and challenging problems in the data mining literature. The problem offers greater flexibility to a decision maker in using item utilities such as profits and margins to mine interesting and actionable patterns from databases. Most of the current works in the literature, however, apply a single minimum utility threshold value and fail to consider disparities in item characteristics. This paper proposes an efficient method (MHUI) to mine high utility itemsets with multiple minimum utility threshold values. The presented method generates high utility itemsets in a single phase without an expensive intermediate candidate generation process. It introduces the concept of suffix minimum utility and presents generalized pruning strategies for efficiently mining high utility itemsets. The performance of the algorithm is evaluated against the state-of-the-art methods (HUI-MMU-TE and HIMU-EUCP) on eight benchmark datasets. The experimental results show that the proposed method delivers two to three orders of magnitude execution time improvement over the HUI-MMU-TE method. In addition, MHUI delivers one to two orders of magnitude execution time improvement over the HIMU-EUCP method, especially on moderately long and dense benchmark datasets. The memory requirements of the proposed algorithm was also found to be significantly lower.

Approximate Parallel High Utility Itemset Mining

High utility itemset mining discovers itemsets whose utility is above a given threshold, where the utility measures the importance of an itemset. It overcomes the limitation of frequent pattern mining, which uses frequency as its quality measure. To speed up the performance for mining high utility itemsets, many algorithms have been proposed which usually focus on optimizing the candidate generation process. However, memory and time performance limitations still cause scalability issues, especially when the dataset is very large. In this paper, the problem is addressed by proposing a distributed parallel algorithm, PHUI-Miner, and a sampling strategy, which can be used either separately or simultaneously. PHUI-Miner parallelizes the state-of-the-art high utility itemset mining algorithm HUI-Miner. In PHUI-Miner, the search space of the high utility itemset mining problem is divided and assigned to nodes in a cluster, which splits the workload. The sampling strategy investigates the required sample size of a dataset, in order to achieve a given accuracy. The sample size is selected based on a new theorem, which provides a theoretical guarantee on the accuracy of results. We also propose an approach combining sampling with PHUI-Miner, which mines an approximate set of results, but could provide better time performance. In our experiments, we show that PHUI-Miner has high performance on different datasets and outperforms the state-of-the-art non-parallel algorithm HUI-Miner. The sampling strategy achieves accuracies much higher than the guarantee provided by the theorems in practice. Extensive experiments are also conducted to compare the time performance of PHUI-Miner with and without sampling.

Mining Spatial Association Rules to Automatic Grouping of Spatial Data Objects Using Multiple Kernel-Based Probabilistic Clustering

AbstractWith the extensive application of spatial databases to various fields ranging from remote sensing to geographical information systems, computer cartography, environmental assessment, and planning, discovery of interesting and hidden knowledge in the spatial databases is a considerable chore for classifying and using the spatial data and knowledge bases. The literature presents different spatial data mining methods to mine knowledge from spatial databases. In this paper, spatial association rules are mined to automatic grouping of spatial data objects using a candidate generation process with three constraint measures, such as support, confidence, and lift. Then, the proposed multiple kernel-based probabilistic clustering is applied to the associate vector to further group the spatial data objects. Here, membership probability based on probabilistic distance is used with multiple kernels, where exponential and tangential kernel functions are utilized. The performance of the proposed method is analyzed with three data sets of three different geometry types using the number of rules and clustering accuracy. From the experimentation, the results proved that the proposed multi-kernel probabilistic clustering algorithm achieved better accuracy as compared with the existing probabilistic clustering.

An Efficient Algorithm for Mining Coherent Association Rules

are many data mining techniques for finding association rules with the predefined minimum support from transaction databases. However, common problems with existing approaches are that an appropriate minimum support is difficult to determine and that the derived rules convey common-sense knowledge. Mining association rules without minimum support threshold is an important approximation of the association rule mining problem, which has been recently proposed(1).In this approach rules are generated based on logical implications and hence fewer and more meaningful rules called coherent rules are reported. An algorithm named ChSearch, is proposed to mine coherent rules for the user specified itemset. In this paper, the same problem is studied with two additional improvements. First, Apriori-based algorithm, namely ECARM for mining all coherent rules is proposed. Second to speed up time consuming candidate generation process an efficient pruning strategy is introduced. The proposed method generates coherent rules in shorter average execution times and fewer database scans, and thereby reduce great amount of I/O time per database scan and hence scaled to large databases. The method has been evaluated using both synthetic and real datasets, and the experimental results demonstrate its effectiveness and efficiency.

HUC-Prune: an efficient candidate pruning technique to mine high utility patterns

Traditional frequent pattern mining methods consider an equal profit/weight for all items and only binary occurrences (0/1) of the items in transactions. High utility pattern mining becomes a very important research issue in data mining by considering the non-binary frequency values of items in transactions and different profit values for each item. However, most of the existing high utility pattern mining algorithms suffer in the level-wise candidate generation-and-test problem and generate too many candidate patterns. Moreover, they need several database scans which are directly dependent on the maximum candidate length. In this paper, we present a novel tree-based candidate pruning technique, called HUC-Prune (High Utility Candidates Prune), to solve these problems. Our technique uses a novel tree structure, called HUC-tree (High Utility Candidates tree), to capture important utility information of the candidate patterns. HUC-Prune avoids the level-wise candidate generation process by adopting a pattern growth approach. In contrast to the existing algorithms, its number of database scans is completely independent of the maximum candidate length. Extensive experimental results show that our algorithm is very efficient for high utility pattern mining and it outperforms the existing algorithms.

Finding frequent itemsets over online data streams

Conventional data mining methods for finding frequent itemsets require considerable computing time to produce their results from a large data set. Due to this reason, it is almost impossible to apply them to an analysis task in an online data stream where a new transaction is continuously generated at a rapid rate. An algorithm for finding frequent itemsets over an online data stream should support flexible trade-off between processing time and mining accuracy. Furthermore, the most up-to-date resulting set of frequent itemsets should be available quickly at any moment. To satisfy these requirements, this paper proposes a data mining method for finding frequent itemsets over an online data stream. The proposed method examines each transaction one-by-one without any candidate generation process. The count of an itemset that appears in each transaction is monitored by a lexicographic tree resided in main memory. The current set of monitored itemsets in an online data stream is minimized by two major operations: delayed-insertion and pruning. The former is delaying the insertion of a new itemset in recent transactions until the itemset becomes significant enough to be monitored. The latter is pruning a monitored itemset when the itemset turns out to be insignificant. The number of monitored itemsets can be flexibly controlled by the thresholds of these two operations. As the number of monitored itemsets is decreased, frequent itemsets in the online data stream are more rapidly traced while they are less accurate. The performance of the proposed method is analyzed through a series of experiments in order to identify its various characteristics.