Gary M. Weiss

Activity recognition using cell phone accelerometers

Mobile devices are becoming increasingly sophisticated and the latest generation of smart cell phones now incorporates many diverse and powerful sensors. These sensors include GPS sensors, vision sensors (i.e., cameras), audio sensors (i.e., microphones), light sensors, temperature sensors, direction sensors (i.e., magnetic compasses), and acceleration sensors (i.e., accelerometers). The availability of these sensors in mass-marketed communication devices creates exciting new opportunities for data mining and data mining applications. In this paper we describe and evaluate a system that uses phone-based accelerometers to perform activity recognition, a task which involves identifying the physical activity a user is performing. To implement our system we collected labeled accelerometer data from twenty-nine users as they performed daily activities such as walking, jogging, climbing stairs, sitting, and standing, and then aggregated this time series data into examples that summarize the user activity over 10- second intervals. We then used the resulting training data to induce a predictive model for activity recognition. This work is significant because the activity recognition model permits us to gain useful knowledge about the habits of millions of users passively---just by having them carry cell phones in their pockets. Our work has a wide range of applications, including automatic customization of the mobile devices behavior based upon a users activity (e.g., sending calls directly to voicemail if a user is jogging) and generating a daily/weekly activity profile to determine if a user (perhaps an obese child) is performing a healthy amount of exercise.

Mining with rarity: a unifying framework

Rare objects are often of great interest and great value. Until recently, however, rarity has not received much attention in the context of data mining. Now, as increasingly complex real-world problems are addressed, rarity, and the related problem of imbalanced data, are taking center stage. This article discusses the role that rare classes and rare cases play in data mining. The problems that can result from these two forms of rarity are described in detail, as are methods for addressing these problems. These descriptions utilize examples from existing research. So that this article provides a good survey of the literature on rarity in data mining. This article also demonstrates that rare classes and rare cases are very similar phenomena---both forms of rarity are shown to cause similar problems during data mining and benefit from the same remediation methods.

Cell phone-based biometric identification

Mobile devices are becoming increasingly sophisticated and now incorporate many diverse and powerful sensors. The latest generation of smart phones is especially laden with sensors, including GPS sensors, vision sensors (cameras), audio sensors (microphones), light sensors, temperature sensors, direction sensors (compasses), and acceleration sensors. In this paper we describe and evaluate a system that uses phone-based acceleration sensors, called accelerometers, to identify and authenticate cell phone users. This form of behavioral biométrie identification is possible because a persons movements form a unique signature and this is reflected in the accelerometer data that they generate. To implement our system we collected accelerometer data from thirty-six users as they performed normal daily activities such as walking, jogging, and climbing stairs, aggregated this time series data into examples, and then applied standard classification algorithms to the resulting data to generate predictive models. These models either predict the identity of the individual from the set of thirty-six users, a task we call user identification, or predict whether (or not) the user is a specific user, a task we call user authentication. This work is notable because it enables identification and authentication to occur unobtrusively, without the users taking any extra actions-all they need to do is carry their cell phones. There are many uses for this work. For example, in environments where sharing may take place, our work can be used to automatically customize a mobile device to a user. It can also be used to provide device security by enabling usage for only specific users and can provide an extra level of identity verification.

Does cost-sensitive learning beat sampling for classifying rare classes?

A highly-skewed class distribution usually causes the learned classifier to predict the majority class much more often than the minority class. This is a consequence of the fact that most classifiers are designed to maximize accuracy. In many instances, such as for medical diagnosis, the minority class is the class of primary interest and hence this classification behavior is unacceptable. In this paper, we compare two basic strategies for dealing with data that has a skewed class distribution and non-uniform misclassification costs. One strategy is based on cost-sensitive learning while the other strategy employs sampling to create a more balanced class distribution in the training set. We compare two sampling techniques, up-sampling and down-sampling, to the cost-sensitive learning approach. The purpose of this paper is to determine which technique produces the best overall classifier---and under what circumstances.

Data Mining in Telecommunications

Telecommunication companies generate a tremendous amount of data. These data include call detail data, which describes the calls that traverse the telecommunication networks, network data, which describes the state of the hardware and software components in the network, and customer data, which decsribes the telecommmunication customers. This chapter describes how Data Mining can be used to uncover useful information buried within these data sets. Several Data Mining applications are described and together they demonstrate that Data Mining can be used to identify telecommunication fraud, improve marketing effectiveness, and identify network faults.

Applications of mobile activity recognition

Activity Recognition (AR), which identifies the activity that a user performs, is attracting a tremendous amount of attention, especially with the recent explosion of smart mobile devices. These ubiquitous mobile devices, most notably but not exclusively smartphones, provide the sensors, processing, and communication capabilities that enable the development of diverse and innovative activity recognition-based applications. However, although there has been a great deal of research into activity recognition, surprisingly little practical work has been done in the area of applications in mobile devices. In this paper we describe and categorize a variety of activity recognition-based applications. Our hope is that this work will encourage the development of such applications and also influence the direction of activity recognition research.

The Impact of Small Disjuncts on Classifier Learning

Many classifier induction systems express the induced classifier in terms of a disjunctive description. Small disjuncts are those that classify few training examples. These disjuncts are interesting because they are known to have a much higher error rate than large disjuncts and are responsible for many, if not most, of all classification errors. Previous research has investigated this phenomenon by performing ad hoc analyses of a small number of data sets. In this chapter we provide a much more systematic study of small disjuncts and analyze how they affect classifiers induced from 30 real-world data sets. A new metric, error concentration, is used to show that for these 30 data sets classification errors are often heavily concentrated toward the smaller disjuncts. Various factors, including pruning, training set size, noise, and class imbalance are then analyzed to determine how they affect small disjuncts and the distribution of errors across disjuncts. This analysis provides many insights into why some data sets are difficult to learn from and also provides a better understanding of classifier learning in general.We believe that such an understanding is critical to the development of improved classifier induction algorithms.

Design considerations for the WISDM smart phone-based sensor mining architecture

Smart phones comprise a large and rapidly growing market. These devices provide unprecedented opportunities for sensor mining since they include a large variety of sensors, including an: acceleration sensor (accelerometer), location sensor (GPS), direction sensor (compass), audio sensor (microphone), image sensor (camera), proximity sensor, light sensor, and temperature sensor. Combined with the ubiquity and portability of these devices, these sensors provide us with an unprecedented view into peoples lives---and an excellent opportunity for data mining. But there are obstacles to sensor mining applications, due to the severe resource limitations (e.g., power, memory, bandwidth) faced by mobile devices. In this paper we discuss these limitations, their impact, and propose a solution based on our WISDM (WIireless Sensor Data Mining) smart phone-based sensor mining architecture.

Maximizing classifier utility when there are data acquisition and modeling costs

Classification is a well-studied problem in data mining. Classification performance was originally gauged almost exclusively using predictive accuracy, but as work in the field progressed, more sophisticated measures of classifier utility that better represented the value of the induced knowledge were introduced. Nonetheless, most work still ignored the cost of acquiring training examples, even though this cost impacts the total utility of the data mining process. In this article we analyze the relationship between the number of acquired training examples and the utility of the data mining process and, given the necessary cost information, we determine the number of training examples that yields the optimum overall performance. We then extend this analysis to include the cost of model induction—measured in terms of the CPU time required to generate the model. While our cost model does not take into account all possible costs, our analysis provides some useful insights and a template for future analyses using more sophisticated cost models. Because our analysis is based on experiments that acquire the full set of training examples, it cannot directly be used to find a classifier with optimal or near-optimal total utility. To address this issue we introduce two progressive sampling strategies that are empirically shown to produce classifiers with near-optimal total utility.

Smartwatch-based biometric gait recognition

The advent of commercial smartwatches provides an intriguing new platform for mobile biometrics. Like their smartphone counterparts, these mobile devices can perform gait-based biometric identification because they too contain an accelerometer and a gyroscope. However, smartwatches have several advantages over smartphones for biometric identification because users almost always wear their watch in the same location and orientation. This location (i.e. the wrist) tends to provide more information about a users movements than the most common location for smartphones (pockets or handbags). In this paper we show the feasibility of using smartwatches for gait-based biometrics by demonstrating the high levels of accuracy that can result from smartwatch-based identification and authentication models. Applications of smartwatch-based biometrics range from a new authentication challenge for use in a multifactor authentication system to automatic personalization by identifying the user of a shared device.