Section: New Results

Data Mining

Participants : Sid Ahmed Benabderrahmane, Marie-Odile Cordier, Serge Vladimir Emteu Tchagou, Thomas Guyet, Yves Moinard, René Quiniou, Alexandre Termier.

Application of sequential pattern mining with intervals

Our theoretical work on sequential pattern mining with intervals [47] has been applied to two real issues: the customer relationship management and analysis of care pathways.

Customer Relationship Management (CRM) comprises a set of tools for managing the interactions between a company and its customers. The main objective of the data analysts is to propose the correct service to a customer at the correct moment by applying decision rules. If rules or sequential patterns can predict the interaction that can follow a sequence of actions or events, they can not predict at what time such actions have the highest probability to occur. The objective of temporal pattern mining is to refine the prediction by extracting patterns with information about the duration and delay between the events. This year we have experimented two algorithms on a CRM databases, QTIPrefixSpan [47] and TGSP [68] , to extract sequential patterns with quantitative temporal information. We have integrated the TGSP algorithm into an interface to visualize and to browse the extracted patterns. A paper describing this contribution have been recently accepted in a workshop [41] .

The QTIPrefixSpan algorithm has also been applied to the analysis of care pathways. The pharmaco-epidemiology platform of the Rennes hospital was interested in characterizing the care pathways preceeding the epileptic seizures of stable epilepic patients. A care pathway consist of the sequence of drug exposures (temporal intervals). The objective is to study the ability of QTIPrefixSpan to identify drug switches between original and generic anti-epileptic drugs. This work is still in progress and will be extended in the PEPS project (see section 8.1.1 ).

Multiscale segmentation of satellite image time series

Satellite images allow the acquisition of large-scale ground vegetation. Images are available along several years with a high acquisition frequency (1 image every two weeks). Such data are called satellite image time series (SITS). In [45] , we presented a method to segment an image through the characterization of the evolution of a vegetation index (NDVI) on two scales: annual and multi-year. This work is now under submission to the journal on Remote Sensing in Environment. The main issue of this approach was the required computation resources (time and memory). Last year, we applied 1D-SAX to reduce data dimensionality [21] . This approach on the supervised classification of large SITS of Senegal and we showed that 1D-SAX approaches the classification results of time series while significantly reducing the required memory storage of the images.

This year, we first continued to explore the supervised classification of SITS using classification trees for time-series [40] by implementing a parallelized version of this algorithm. Secondly, we explored the adaption of the object-oriented segmentation to SITS. The object-oriented segmentation [34] is able to segment images based on the segment uniformity. We proposed a measure for time-series uniformity to adapt the segmentation algorithm and applied it on large multivariate SITS of Senegal. This work have been presented to the conference on spatial analysis and geography [16] . A collaboration with A. Fall (Université Paris-13) have been initiated to compare our results on the Senegal with ground observations. Moreover, we planned to apply our algorithm to analyse the land use in Peru (collaboration with A. Marshall, Université Paris 13/PRODIG).

Analysis and simulation of landscape based on spatial patterns

Researchers in agro-environment need a great variety of landscapes to test their scientific hypotheses using agro-ecological models. Real landscapes are difficult to acquire and do not enable the agronomist to test all their hypothesis. Working with simulated landscapes is then an alternative to get a sufficient variety of experimental data. Our objective is to develop an original scheme to generate landscapes that reproduce realistic interface properties between parcels. This approach is made of the extraction of spatial patterns from a real geographic area and the use of these patterns to generate new "realistic" landscapes. It is based on a spatial representation of landscapes by a graph expressing the spatial relationships between the agricultural parcels (as well as the roads, the rivers, the buildings, etc.), of a specific geographic area.

In past years, we worked on the exploration of graph mining techniques, such as gSPAN [67] , to discover the relevant spatial patterns present in a spatial-graph. We assume that the set of the frequent graph patterns are the characterisation of the landscape. Our remaining challenge was to simulate new realistic landscapes that will reproduce the same patterns.

This year, we formalized the simulation process by a formal problem of graph packing [51] . The process is illustrated by Figure 1 . Solving instances of the general graph packing problem has a high combinatorics and there does not exists any efficient algorithm to solve it. We proposed an ASP program to tackle the combinatorics of the graph packing and to assign the land use considering some expert knowledge. Our approach combines the efficiency of ASP to solve the packing issue and the simplicity of the declarative programming to take into account the expert contraints on the land use. Contraints about the minimum surface of crops or about the impossibility of some crops colocation can be easily defined. This work have been presented at the conference RFIA [19] and we have been invited to provide an extended version to the Revue d'Intelligence Artificielle (RIA). The application results have been presented to the national colloquium on landscape modelling (http://www.reseau-payote.fr/?q=colloque2014 ).

In addition to the landscape simulation, the challenging tasks of solving the general graph packing with ASP raises interests in more general problem (such as graph compression). We initiated a collaboration with J. Nicolas (Inria/Dyliss) to improve the efficiency of our first programs.

Mining with ASP

In pattern mining, a pattern is considered interesting if it occurs frequently in the data, i.e. the number of its occurrences is greater than a fixed given threshold. As non informed mining methods tend to generate massive results, there is more and more interest in pattern mining algorithms able to mine data considering some expert knowledge. Though a generic pattern mining tool that could be tailored to the specific task of a data-scientist is still a holy grail for pattern mining software designers, some recent attempts have proposed generic pattern mining tools [44] for itemset mining tasks. In collaboration with Torsten Schaub, we explore the ability of a declarative language, such as Answer Set Programming (ASP), to solve pattern mining tasks efficiently. A first attempt have been proposed by Jarvisälo for simple settings [49] .

This year, we worked on several classical pattern mining tasks: episodes, sequences and closed/maximal itemsets. In [20] , we explore the use of ASP to extract frequent episodes (without parallel events) in a unique long sequence of itemsets. We especially evaluate the incremental resolution to improve the efficiency of our program. We next worked on sequence mining to extract pattern from the sequence of TV programs (V. Claveau, CNRS/LinkMedia). This tasks was simpler, but the computation time was significantly higher than dedicated algorithms. Nonetheless, our recent programs extracting closed or maximal patterns have better results.

Monitoring cattle

Following the lines of a previous work [62] , we are working on a method for detecting Bovine Respiratory Diseases (BRD) from behavioral (walking, lying, feeding and drinking activity) and physiological (rumen temperature) data recorded on feedlot cattle being fattened up in big farms in Alberta (Canada). This year, we have especially worked on multivariate sensor analysis to devising multivariate decision rules for improving the specificity of detectors [15] .

Subdimensional clustering for fast similarity search over time series data. Application to Information retrieval tasks

Information retrieval and similarity search tasks in time series databases remains a challenge that requires to discover relevant pattern-sequences that are recurrent over the overall time series sequences, and to find temporal associations among these frequently occurring patterns. However, proposed methods suffer from a lack of flexibility of the used similarity measures, a lack of scalability of the representation model, and a penalizing runtime to retrieve the information. Motivated by these observations, we have designed a framework tackling the query by content problem on time series data, ensuring (i) fast response time, (ii) multi-level information representation, and (iii) representing temporal associations between extracted patterns. This year we have compared several distance measures on time series with different criteria and proposed a hybrid retrieval method based on pattern extraction and clustering [8] .

Knowledge Extraction from Heterogeneous Data

Recently, mining microarrays data has become a big challenge due to the growing sources of available data. We are using machine learning methods such as clustering, dimensionality reduction, association rules discovery on transcriptomic data, by combining a domain ontology as source of knowledge, in order to supervise the KDD process. Our objectives concern the identification of genes that could participate in the development of tumors. This year, we have introduced a new method for extracting enriched biological functions from transcriptomic databases using an integrative bi-classication approach based on formal concept analysis [7] .

Trace reduction

One problem of execution trace of applications on embedded systems is that they can grow very large, typically several Gigabytes for 5 minutes of audio/video playback. Some endurance tests require continuous playback for 96 hours, which would lead to hundreds of Gigabytes of traces, that current techniques cannot analyze. We have proposed TraceSquiz, an online approach to monitor the trace output during endurance test, in order to record only suspicious portions of the trace and discard regular ones. This approach is based on anomaly detection techniques, and as been accepted in the DATE'15 conference [14] .