Sergio Di Martino

A Probabilistic Based Approach towards Software System Clustering

In this paper we present a clustering based approach to partition software systems into meaningful subsystems. In particular, the approach uses lexical information extracted from four zones in Java classes, which may provide a different contribution towards software systems partitioning. To automatically weigh these zones, we introduced a probabilistic model, and applied the Expectation-Maximization (EM) algorithm. To group classes according to the considered lexical information, we customized the well-known K-Medoids algorithm. To assess the approach and the implemented supporting system, we have conducted a case study on six open source software systems.

A genetic algorithm to configure support vector machines for predicting fault-prone components

In some studies, Support Vector Machines (SVMs) have been turned out to be promising for predicting fault-prone software components. Nevertheless, the performance of the method depends on the setting of some parameters. To address this issue, we propose the use of a Genetic Algorithm (GA) to search for a suitable configuration of SVMs parameters that allows us to obtain optimal prediction performance. The approach has been assessed carrying out an empirical analysis based on jEdit data from the PROMISE repository. We analyzed both the inter- and the intra-release performance of the proposed method. As benchmarks we exploited SVMs with Grid-search and several other machine learning techniques. The results show that the proposed approach let us to obtain an improvement of the performance with an increasing of the Recall measure without worsening the Precision one. This behavior was especially remarkable for the inter-release use with respect to the other prediction techniques.

Investigating the use of Support Vector Regression for web effort estimation

Support Vector Regression (SVR) is a new generation of Machine Learning algorithms, suitable for predictive data modeling problems. The objective of this paper is twofold: first, to investigate the effectiveness of SVR for Web effort estimation using a cross-company dataset; second, to compare different SVR configurations looking at the one that presents the best performance. In particular, we took into account three variables’ preprocessing strategies (no-preprocessing, normalization, and logarithmic), in combination with two different dependent variables (effort and inverse effort). As a result, SVR was applied using six different data configurations. Moreover, to understand the suitability of kernel functions to handle non-linear problems, SVR was applied without a kernel, and in combination with the Radial Basis Function (RBF) and the Polynomial kernels, thus obtaining 18 different SVR configurations. To identify, for each configuration, which were the best values for each of the parameters we defined a procedure based on a leave-one-out cross-validation approach. The dataset employed was the Tukutuku database, which has been adopted in many previous Web effort estimation studies. Three different training and test set splits were used, including respectively 130 and 65 projects. The SVR-based predictions were also benchmarked against predictions obtained using Manual StepWise Regression and Case-Based Reasoning. Our results showed that the configuration corresponding to the logarithmic features’ preprocessing, in combination with the RBF kernel provided the best results for all three data splits. In addition, SVR provided significantly superior prediction accuracy than all the considered benchmarking techniques.

Integrating Google Earth within OLAP tools for Multidimensional Exploration and Analysis of Spatial Data

Spatial OnLine Analytical Processing solutions are a type of Business Information Tool meant to support a Decision Maker in extracting hidden knowledge from data warehouses containing spatial data. To date, very few SOLAP tools are available, each presenting some drawbacks reducing their flexibility. To overcome these limitations, we have developed a web-based SOLAP tool, obtained by suitably integrating into an ad-hoc architecture the Geobrowser Google Earth with a freely available OLAP engine, namely Mondrian. As a consequence, a Decision Maker can perform exploration and analysis of spatial data both through the Geobrowser and a Pivot Table in a seamlessly fashion. In this paper, we illustrate the main features of the system we have developed, together with the underlying architecture, using a simulated case study.

Handy: A New Interaction Device for Vehicular Information Systems

The design of interfaces for automotive information systems is a critical task. In fact, such design must take into account that user is busy in the primary driving task, and any visual distraction determined by telematics systems can cause serious safety problems. To limit such distraction and enhance safety, in this paper we propose a novel multimodal user interface. The key element of the proposal is a new interaction device, named Handy, conceived to exploit the driver’s tactile channel to minimize the workload of visual channel. Moreover Handy is suitably integrated with the graphical user interface, which is characterized by a reduced number of choices for each state and has been designed in agreement with the self-revealing approach.

Association of Arginine Vasopressin-Secreting Cell, Steroid-Secreting Cell, Adrenal and Islet Cell Antibodies in a Patient Presenting with Central Diabetes insipidus, Empty Sella, Subclinical Adrenocortical Failure and Impaired Glucose Tolerance

A 36-year-old woman with central diabetes insipidus (DI), diagnosed when she was 7, was referred to our Endocrine Unit in January 1993 for further hormonal investigations. Clinical and laboratory findings confirmed the diagnosis of central DI. Cranial computed tomography and magnetic resonance imaging showed only an empty sella. Moreover, we noted impaired glucose tolerance and unusual findings of subclinical adrenocortical failure, i.e. high plasma renin activity with normal aldosterone levels, high ACTH despite normal basal and ACTH-stimulated cortisol levels. Immunological study of the patients serum showed the presence of arginine vasopressin (AVP)-secreting cell antibodies (Abs), steroid-producing cell Abs, adrenal and islet cell Abs. The following aspects of our case are stressed and discussed: (1) the presence of AVP-secreting cell Abs 29 years after the diagnosis of DI; (2) the association between DI, empty sella and subclinical autoimmune adrenocortical failure with unusual hormonal findings, and (3) impaired glucose tolerance with islet cell antibody positivity.

Pregnancy may favour the development of severe autoimmune central diabetes insipidus in women with vasopressin cell antibodies: description of two cases

Recently, an increased incidence of central diabetes insipidus (CDI) in pregnancy, and less frequently in the post partum period, has been reported, most probably favoured by some conditions occurring in pregnancy. This study was aimed at investigating the influence of pregnancy on a pre-existing potential/subclinical hypothalamic autoimmunity. We studied the longitudinal behaviour of arginine-vasopressin cell antibodies (AVPcAbs) and post-pituitary function in two young women with a positive history of autoimmune disease and presence of AVPcAbs, but without clinical CDI, and who became pregnant 5 and 7 months after our first observation. The behaviour of post-pituitary function and AVPcAbs (by immunofluorescence) was evaluated at baseline, during pregnancy and for 2 years after delivery. AVPcAbs, present at low/middle titres at baseline in both patients, showed a titre increase during pregnancy in one patient and after delivery in the other patient, with development of clinically overt CDI. Therapy with 1-deamino-8-d-arginine vasopressin (DDAVP) caused a prompt clinical remission. After a first unsuccessful attempt of withdrawal, the therapy was definitively stopped at the 6th and the 7th month of post partum period respectively, when AVPcAbs disappeared, accompanied by post-pituitary function recovery, persisting until the end of the follow-up. The determination of AVPcAbs is advisable in patients with autoimmune diseases planning their pregnancy, because they could be considered good predictive markers of gestational or post partum autoimmune CDI. The monitoring of AVPcAb titres and post-pituitary function during pregnancy in these patients may allow for an early diagnosis and an early replacement therapy, which could induce the disappearance of these antibodies with consequent complete remission of CDI.

Web Effort Estimation

Context: software development effort estimation is a crucial management task that critically depends on the adopted size measure. Several Functional Size Measurement (FSM) methods have been proposed. COSMIC is considered a 2nd generation FSM method, to differentiate it from Function Point Analysis (FPA) and its variants, considered as 1st generation ones. In the context of Web applications, few investigations have been performed to compare the effectiveness of the two generations. Software companies could benefit from this analysis to evaluate if it is worth to migrate from a 1st generation method to a 2nd one.Objective: the main goal of the paper is to empirically investigate if COSMIC is more effective than FPA for Web effort estimation. Since software companies using FPA cannot build an estimation model based on COSMIC as long as they do not have enough COSMIC data, the second goal of the paper is to investigate if conversion equations can be exploited to support the migration from FPA to COSMIC.Method: two empirical studies have been carried out by employing an industrial data set. The first one compared the effort prediction accuracy obtained with Function Points (FPs) and COSMIC, using two estimation techniques (Simple Linear Regression and Case-Based Reasoning). The second study assessed the effectiveness of a two-step strategy that first exploits a conversion equation to transform historical FPs data into COSMIC, and then builds a new prediction model based on those estimated COSMIC sizes.Results: the first study revealed that, on our data set, COSMIC was significantly more accurate than FPs in estimating the development effort. The second study revealed that the effectiveness of the analyzed two-step process critically depends on the employed conversion equation.Conclusion: for Web effort estimation COSMIC can be significantly more effective than FPA. Nevertheless, additional research must be conducted to identify suitable conversion equations so that the two-step strategy can be effectively employed for a smooth migration from FPA to COSMIC.

Weighing lexical information for software clustering in the context of architecture recovery

In this paper, we present a software clustering approach that leverages the information conveyed by the zone in which each lexeme appears in the classes of object oriented systems. We define six zones in the source code: Class Name, Attribute Name, Method Name, Parameter Name, Comment, and Source Code Statement. These zones may convey information with different levels of relevance, and so their contribution should be differently weighed according to the software system under study. To this aim, we define a probabilistic model of the lexemes distribution whose parameters are automatically estimated by the Expectation-Maximization algorithm. The weights of the zones are then exploited to compute similarities among source code classes, which are then grouped by a k-Medoid clustering algorithm. To assess the validity of our solution in the software architecture recovery field, we applied our approach to 19 software systems from different application domains. We observed that the use of our probabilistic model and the defined zones improves the quality of clustering results so that they are close to a theoretical upper bound we have proved.

What are the potentialities of crowdsourcing for dynamic maps of on-street parking spaces?

Finding a parking space is a crucial mobility problem, which could be mitigated by dynamic maps of parking availability. The creation of these maps requires current information on the state of the parking stalls, which could be obtained by (I) instrumenting the road infrastructure with sensors, (II) using probe vehicles, or (III) using mobile apps. In this paper, we investigate the potential predictive performances of a random forest binary classifier, comparing these three data collection strategies. As for the dataset, we used real infrastructure measurements in San Francisco for solution I. We simulated the crowdsourcing solutions II and III by downsampling that dataset, based on different assumptions. Evaluations show that the instrumented solution is clearly superior over the two crowdsourcing strategies, but with remarkably small differences to the probe vehicle scenario. On the other hand, a mobile app would require a very high penetration rate in order to be used for meaningful predictions.