Gabriella Schoier

Density Analysis on Large Geographical Databases. Search for an Index of Centrality of Services at Urban Scale

Geographical databases are available to date containing detailed and georeferenced data on population, commercial activities, business, transport and services at urban level. Such data allow examining urban phenomena at very detailed scale but also require new methods for analysis, comprehension and visualization of the spatial phenomena. In this paper a density-based method for extracting spatial information from large geographical databases is examined and first results of its application at the urban scale are presented. Kernel Density Estimation is used as a density based technique to detect clusters in spatial data distributions. GIS and spatial analytical methods are examined to detect areas of high services’ supply in an urban environment. The analysis aims at identifying clusters of services in the urban environment and at verifying the correspondence between urban centres and high levels of service.

Using clustering for document reconstruction

In the forensics and investigative science fields there may arise the need of reconstructing documents which have been destroyed by means of a shredder. In a computer-based reconstruction, the pieces are described by numerical features, which represent the visual content of the strips. Usually, the pieces of different pages have been mixed. We propose an approach for the reconstruction which performs a first clustering on the strips to ease the successive matching, be it manual (with the help of a computer) or automatic. A number of features, extracted by means of image processing algorithms, have been selected for this aim. The results show the effectiveness of the features and of the proposed clustering algorithm.

A Clustering Method for Large Spatial Databases

The rapid developments in the availability and access to spatially referenced information in a variety of areas, has induced the need for better analysis techniques to understand the various phenomena. In particular spatial clustering algorithms which groups similar spatial objects into classes can be used for the identification of areas sharing common characteristics. The aim of this paper is to present a density-based algorithm for the discover of clusters in large spatial data set which is a modification of a recently proposed algorithm.This is applied to a real data set related to homogeneous agricultural environments.

A methodology for dealing with spatial big data

Spatial data mining (SDM) refers to the mining of knowledge from spatial data. Recently, location-based services have enabled the gathering of a significant amount of geo-referenced data, i.e., of spatial big data (SBD). Spatial datasets often exceed the ability of current computing systems to manage these data with reasonable effort; therefore, data-intensive computing and data mining techniques are useful tools for conducting an analysis. In this paper, we present an approach to the clustering of high-dimensional data that allows a flexible approach to the statistical modelling of phenomena characterised by unobserved heterogeneity. Numerous clustering algorithms have been developed for large databases; density-based algorithms particularly treat a huge amount of data in large spatial databases. We present the Modified Density-Based Spatial Clustering of Applications with Noise (MDBSCAN) algorithm and compare it to the classical k-means approach. Both applications use synthetic datasets and a dataset of satellite images.

A Different Approach for the Analysis of Web Access Logs

The development of Internet-based business has pointed out the importance of the personalisation and optimisation of Web sites. For this purpose the study of users behaviours are of great importance. In this paper we present a solution to the problem of identification of dense clusters in the analysis of Web Access Logs. We consider a modification of an algorithm recently proposed in social network analysis. This approach is illustrated by analysing a log-file of a web portal.

Blockmodeling Techniques for Web Mining

The aim of this paper is to introduce the concept of blockmodeling to Web data (log files), in order to obtain clusters of units (visited pages) which have similar patterns of relationships and to interpret the pattern of relationships among clusters. Doing this we can study the behaviour of the users (sessions) according to the relations among the visited pages. To explain these concepts an application to a Web site is performed.

On the Problem of Clustering Spatial Big Data

Different motivation are related with the analysis of Spatial Big Data (SBD). Google Earth, Google Maps, Navigation, location-based service allow to obtain a great amount of geo-referenced data. Often spatial datasets exceed the capacity of current computing systems to manage, process, or analyze the data with reasonable effort. Considering SBD history methodology as Data-intensive Computing and Data Mining techniques have been useful. In this context the problem regards the analysis of of high frequency spatial data. In this paper we present an approach to clustering of high dimensional data which allows a flexible approach to the statistical modeling of phenomena characterized by unobserved heterogeneity. We consider the MDBSCAN and compare it with the classical k-means approach. The applications concern a synthetic data set and a data set of satellite images.

SETAR model selection-A bootstrap approach

SummaryThe aim of this paper is to propose new selection criteria for the orders of selfexciting threshold autoregressive (SETAR) models. These criteria use bootstrap methodology; they are based on a weighted mean of the apparent error rate in the sample and the average error rate obtained from bootstrap samples not containing the point being predicted. These new criteria are compared with the traditional ones based on the Akaike information criterion (AIC). A simulation study and an example on a real data set end the paper.

Strong consistency of conditional least squares estimators in multiple regime threshold autoregressive models

In this paper we analyse the conditional least squares estimators of the parameters of a multiple regime threshold AR(1) model and prove that under certain conditions these are strongly consistent. We assume that the error process in each regime is amartingale difference sequence. Then we deal with strong consistency of the natural estimator of the error variance in each regime.

Clustering Algorithms for Spatial Big Data

In our time people and devices constantly generate data. User activity generates data about needs and preferences as well as the quality of their experiences in different ways: i.e. streaming a video, looking at the news, searching for a restaurant or a an hotel, playing a game with others, making purchases, driving a car. Even when people put their devices in their pockets, the network is generating location and other data that keeps services running and ready to use. This rapid developments in the availability and access to data and in particular spatially referenced data in a different areas, has induced the need for better analysis techniques to understand the various phenomena. Spatial clustering algorithms, which groups similar spatial objects into classes, can be used for the identification of areas sharing common characteristics. The aim of this paper is to analyze the performance of three different clustering algorithms i.e. the Density-Based Spatial Clustering of Applications with Noise algorithm (DBSCAN), the Fast Search by Density Peak (FSDP) algorithm and the classic K-means algorithm (K-Means) as regards the analysis of spatial big data. We propose a modification of the FSDP algorithm in order to improve its efficiency in large databases. The applications concern both synthetic data sets and satellite images.