Mining Temporal Patterns Research Articles

Self-Attention Pooling-Based Long-Term Temporal Network for Action Recognition

With the development of Internet of Things (IoT), self-driving technology has been successful. Yet safe driving faces challenges due to such cases as pedestrians crossing roads. How to sense their movements and identify their behaviors from video data is important. Most of the existing methods fail to: 1) capture long-term temporal relationship well due to their limited temporal coverage and 2) aggregate discriminative representation effectively, such as caused by little or even no attention paid to differences among representations. To address such issues, this work presents a new architecture called a self-attention pooling-based long-term temporal network (SP-LTN), which can learn long-term temporal representations and aggregate those discriminative representations in an end-to-end manner, and on the other hand, effectively conduct long-term representation learning on a given video by capturing spatial information and mining temporal patterns. Next, it develops a self-attention pooling method to predict the importance scores of obtained representations for distinguishing them from each other and then weights them together to highlight the contributions of those discriminative representations in action recognition. Finally, it designs a new loss function that combines a standard cross-entropy loss function with a regularization term to further focus on the discriminative representations while restraining the impact of distractive ones on activity classification. Experimental results on two data sets show that our SP-LTN, fed by only red–green–blue (RGB) frames, outperforms the state-of-the-art methods.

Efficient temporal pattern mining in big time series using mutual information

Very large time series are increasingly available from an ever wider range of IoT-enabled sensors deployed in different environments. Significant insights can be gained by mining temporal patterns from these time series. Unlike traditional pattern mining, temporal pattern mining (TPM) adds event time intervals into extracted patterns, making them more expressive at the expense of increased time and space complexities. Existing TPM methods either cannot scale to large datasets, or work only on pre-processed temporal events rather than on time series. This paper presents our Frequent Temporal Pattern Mining from Time Series (FTPMfTS) approach providing: (1) The end-to-end FTPMfTS process taking time series as input and producing frequent temporal patterns as output. (2) The efficient Hierarchical Temporal Pattern Graph Mining (HTPGM) algorithm that uses efficient data structures for fast support and confidence computation, and employs effective pruning techniques for significantly faster mining. (3) An approximate version of HTPGM that uses mutual information, a measure of data correlation, to prune unpromising time series from the search space. (4) An extensive experimental evaluation showing that HTPGM outperforms the baselines in runtime and memory consumption, and can scale to big datasets. The approximate HTPGM is up to two orders of magnitude faster and less memory consuming than the baselines, while retaining high accuracy.

Quantitative and temporal approach to utilising electronic medical records from general practices in mental health prediction

This study proposes a framework for mining temporal patterns from Electronic Medical Records. A new scoring scheme based on the Wilson interval is provided to obtain frequent and predictive patterns, as well as to accelerate the mining process by reducing the number of patterns mined. This is combined with a case study using data from general practices in the Netherlands to identify children at risk of suffering from mental disorders. To develop an accurate model, feature engineering methods such as one hot encoding and frequency transformation are proposed, and the pattern selection is tailored to this type of clinical data. Six machine learning models are trained on five age groups, with XGBoost achieving the highest AUC values (0.75–0.79) with sensitivity and specificity above 0.7 and 0.6 respectively. An improvement is demonstrated by the models learning from patterns in addition to non-temporal features.

Temporal Association Rule Mining and Updating and Their Application to Blast Furnace in the Steel Industry.

Blast furnace (BF) is the main method of modern iron-making. Ensuring the stability of the BF conditions can effectively improve the quality and output of iron and steel. However, operations of BF depend on mainly human experience, which causes two problems: (1) human experience is not objective and is difficult to inherit and learn and (2) it is difficult to acquire knowledge that contains time information among multiple variables in BF. To address these problems, a data-driven method is proposed. In this article, we propose a novel and efficient algorithm for discovering underlying knowledge in the form of temporal association rules (TARs) in BF iron-making data. First, a new TAR mining framework is proposed for mining temporal frequent patterns. Then, a novel TAR mining algorithm is proposed for mining underlying, up-to-date, and effective knowledge in the form of TARs. Finally, considering the updating of the BF database, a rule updating method is proposed that is based on the algorithm that is proposed in this article. Our extensive experiments demonstrate the satisfactory performance of the proposed algorithm in discovering TARs in comparison with the state-of-the-art algorithms. Experiments on BF iron-making data have demonstrated the superior performance and practicability of the proposed method.

Mining Skier Transportation Patterns From Ski Resort Lift Usage Data

Descriptive data analysis is used for mining spatial and temporal patterns from ski lift entrance data. The data, collected through radio-frequency identification scanners, cover one skiing season with approximately 1.2 million recorded ski lift transportations. Cluster analysis was performed on ten subsamples, and the obtained clusters were cross-validated. Several types of skier behavior were found. Temporal clustering revealed that skier patterns differ according to time of maximal performance and length of stay in the ski lift transportation system. Spatial clustering revealed that it is reasonable to have as many clusters as ski lifts in a ski resort, since skiers tend to choose a dominant ski lift during a skier-day. The detected patterns reveal valuable information that can be used for potential improvement of products and services offered by ski resorts.

A novel fuzzy gaussian-based dissimilarity measure for discovering similarity temporal association patterns

Mining temporal association patterns from time-stamped temporal databases, first introduced in 2009, remain an active area of research. A pattern is temporally similar when it satisfies certain specified subset constraints. The naive and apriori algorithm designed for non-temporal databases cannot be extended to find similar temporal patterns in the context of temporal databases. The brute force approach requires performing $$2^{n }$$ true support computations for ‘n’ items; hence, an NP-class problem. Also, the apriori or fp-tree-based algorithms designed for static databases are not directly extendable to temporal databases to retrieve temporal patterns similar to a reference prevalence of user interest. This is because the support of patterns violates the monotonicity property in temporal databases. In our case, support is a vector of values and not a single value. In this paper, we present a novel approach to retrieve temporal association patterns whose prevalence values are similar to those of the user specified reference. This allows us to significantly reduce support computations by defining novel expressions to estimate support bounds. The proposed approach eliminates computational overhead in finding similar temporal patterns. We then introduce a novel dissimilarity measure, which is the fuzzy Gaussian-based dissimilarity measure. The measure also holds the monotonicity property. Our evaluations demonstrate that the proposed method outperforms brute force and sequential approaches. We also compare the performance of the proposed approach with the SPAMINE which uses the Euclidean measure. The proposed approach uses monotonicity property to prune temporal patterns without computing unnecessary true supports and distances.

Mining frequent arrangements and sequencing of events for diseases prognosis using reference event-based temporal relations

Mining temporal patterns has been an active research area in AI. Mining patterns of temporally correlated events called arrangements is required in many scientific domains, especially in the medical domain. Key events of the domain defined as reference events are used in describing the arrangements. In this work, mining events in arrangements (MENSA) algorithm is proposed to mine frequent arrangements and to generate sequential rules that can be used in early forecasting of probable diseases and symptoms and hence facilitate decision support related tasks. The MENSA algorithm has excellent scale-up property with respect to the size of the sequences. Temporal relations between events are defined using the contextual reference event-based temporal (REseT) relations. These relations can provide more useful knowledge about the order of events within an arrangement in comparison with Allen's relations and the effectiveness, applicability of these relations in prognosis is demonstrated using PREDICT algorithm with Percent_Similarity as the performance measure. The performance of the proposed algorithms is evaluated and they outperform other existing approaches when experimented on real finance dataset and synthetic medical dataset.

Mining frequent arrangements and sequencing of events for diseases prognosis using reference event-based temporal relations

Mining temporal patterns has been an active research area in AI. Mining patterns of temporally correlated events called arrangements is required in many scientific domains, especially in the medical domain. Key events of the domain defined as reference events are used in describing the arrangements. In this work, mining events in arrangements MENSA algorithm is proposed to mine frequent arrangements and to generate sequential rules that can be used in early forecasting of probable diseases and symptoms and hence facilitate decision support related tasks. The MENSA algorithm has excellent scale-up property with respect to the size of the sequences. Temporal relations between events are defined using the contextual reference event-based temporal REseT relations. These relations can provide more useful knowledge about the order of events within an arrangement in comparison with Allen's relations and the effectiveness, applicability of these relations in prognosis is demonstrated using PREDICT algorithm with Percent_Similarity as the performance measure. The performance of the proposed algorithms is evaluated and they outperform other existing approaches when experimented on real finance dataset and synthetic medical dataset.

Discovering Locally Frequent Patterns from Temporal Data

Mining patterns from large dataset is an interested data mining problem. Many methods have been developed for this purpose till today. Most of the methods considered the time attributes as one of the attributes like others. However taking the time attribute into account separately the patterns can be extracted which cannot be extracted by normal methods. These patterns are termed as temporal patterns A couple of works have already been done in mining temporal patterns. In this article, we propose an efficient method of mining locally frequent patterns from temporal datasets. The efficacy of our method is established using experiments carried on a real life datasets and a synthetic datasets. Then we make comparative studies with a well established method and found that our method outperforms at least one well-known method.

Prediction of Human Activity by Discovering Temporal Sequence Patterns.

Early prediction of ongoing human activity has become more valuable in a large variety of time-critical applications. To build an effective representation for prediction, human activities can be characterized by a complex temporal composition of constituent simple actions and interacting objects. Different from early detection on short-duration simple actions, we propose a novel framework for long -duration complex activity prediction by discovering three key aspects of activity: Causality, Context-cue, and Predictability. The major contributions of our work include: (1) a general framework is proposed to systematically address the problem of complex activity prediction by mining temporal sequence patterns; (2) probabilistic suffix tree (PST) is introduced to model causal relationships between constituent actions, where both large and small order Markov dependencies between action units are captured; (3) the context-cue, especially interactive objects information, is modeled through sequential pattern mining (SPM), where a series of action and object co-occurrence are encoded as a complex symbolic sequence; (4) we also present a predictive accumulative function (PAF) to depict the predictability of each kind of activity. The effectiveness of our approach is evaluated on two experimental scenarios with two data sets for each: action-only prediction and context-aware prediction. Our method achieves superior performance for predicting global activity classes and local action units.

An intelligent neuro fuzzy temporal knowledge representation model for mining temporal patterns

Representation of temporal knowledge and analysis of temporal data is becoming a good practice for effective classification and prediction. Various semantic levels on knowledge representation schemes have been measured for temporal data. The existing Fuzzy Cognitive Maps FCMs facilitate modeling dynamic systems for knowledge representation and reasoning under uncertainty. However, the FCMs are constructed manually and are constrained by the human experts' validation for assessing its reliability and they are lacking in considering temporal features necessary for reasoning in medical applications. This paper proposes a new temporal mining system known as Fuzzy Temporal Cognitive Map FTCM, which defines a complete discrete temporal extension and fuzzy inference mechanism of FCM. In FTCM, the temporal dependencies of concepts during a particular time interval are measured. This work aims to reduce the complexities of dynamic modeling of a complex causal system by proposing a four layer fuzzy neural network to construct FTCM from the temporal data. In this proposed model, a fuzzy temporal mutual subsethood operator is used to measure the activation spread in the FTCM for automatic quantification of causalities. This FTCM is designed for a set of temporal clinical records, which can be further used for inferencing and prediction in medical diagnosis by generating a set of fuzzy temporal rules using Allen's temporal relationships and fuzzy temporal rules.

Mining temporal patterns in popularity of web items

Huge amounts of various web items (e.g., images, keywords, and web pages) are being made available on the Web. The popularity of such web items continuously changes over time, and mining for temporal patterns in the popularity of web items is an important problem that is useful for several Web applications; for example, the temporal patterns in the popularity of web search keywords help web search enterprises predict future popular keywords, thus enabling them to make price decisions when marketing search keywords to advertisers. However, the presence of millions of web items makes it difficult to scale up previous techniques for this problem. This paper proposes an efficient method for mining temporal patterns in the popularity of web items. We treat the popularity of web items as time-series and propose a novel measure, a gap measure, to quantify the dissimilarity between the popularity of two web items. To reduce the computational overhead for this measure, an efficient method using the Discrete Fourier Transform (DFT) is presented. We assume that the popularity of web items is not necessarily periodic. For finding clusters of web items with similar popularity trends, we show the limitations of traditional clustering approaches and propose a scalable, efficient, density-based clustering algorithm using the gap measure. Our experiments using the popularity trends of web search keywords obtained from the Google Trends web site illustrate the scalability and usefulness of the proposed approach in real-world applications.

MILPRIT*

In this article, we consider a new kind of temporal pattern where both interval and punctual time representation are considered. These patterns, which we call temporal point-interval patterns, aim at capturing how events taking place during different time periods or at different time instants relate to each other. The datasets where these kinds of patterns may appear are temporal relational databases whose relations contain point or interval timestamps. We use a simple extension of Allen’s Temporal Interval Logic as a formalism for specifying these temporal patterns. We also present the algorithm MILPRIT* for mining temporal point-interval patterns, which uses variants of the classical levelwise search algorithms. In addition, MILPRIT* allows a broad spectrum of constraints to be incorporated into the mining process. An extensive set of experiments of MILPRIT* executed over synthetic and real data is presented, showing its effectiveness for mining temporal relational patterns.

유사한 인기도 추세를 갖는 웹 객체들의 클러스터링

Huge amounts of various web items such as keywords, images, and web pages are being made widely available on the Web. The popularities of such web items continuously change over time, and mining temporal patterns in popularities of web items is an important problem that is useful for several web applications. For example, the temporal patterns in popularities of search keywords help web search enterprises predict future popular keywords, enabling them to make price decisions when marketing search keywords to advertisers. However, presence of millions of web items makes it difficult to scale up previous techniques for this problem. This paper proposes an efficient method for mining temporal patterns in popularities of web items. We treat the popularities of web items as time-series, and propose gapmeasure to quantify the similarity between the popularities of two web items. To reduce the computation overhead for this measure, an efficient method using the Fast Fourier Transform (FFT) is presented. We assume that the popularities of web items are not necessarily following any probabilistic distribution or periodic. For finding clusters of web items with similar popularity trends, we propose to use a density-based clustering algorithm based on the gap measure. Our experiments using the popularity trends of search keywords obtained from the Google Trends web site illustrate the scalability and usefulness of the proposed approach in real-world applications.