Miguel Felgueiras

Fall Detection on Ambient Assisted Living using a Wireless Sensor Network

In this work, a distributed system for fall detection is presented. The proposed system was designed to monitor activities of the daily living of elderly people and to inform the caregivers when a falls event occurs. This system uses a scalable wireless sensor networks to collect the data and transmit it to a control center. Also, an intelligent algorithm is used to process the data collected by the sensor networks and calculate if an event is, or not, a fall. A statistical method is used to improve this algorithm and to reduce false positives. The system presented has the capability to learn with past events and to adapt is behavior with new information collected from the monitored elders. The results obtained show that the system has an accuracy above 98%.

Known Mean, Unknown Maxima? Testing the Maximum Knowing Only the Mean

In the quantitative group testing problem, the use of the group mean to identify if the group maximum is greater than a prefixed threshold (infected group) is analyzed, using n independent and identically distributed individuals. Under these conditions, it is shown that the information of the mean is sufficient to classify each group as infected or healthy with low probability of misclassification when the underline distribution is a unilateral heavy-tailed distribution.

Pseudo-convex mixtures

Allowing weights w∈]−1;1]\{0}, pseudo-convex mixtures increase usual mixtures flexibility. The finite mixtures with negative components are investigated for distribution families closed under minimization. The main purpose is to define these mixtures and to study their properties.

Elder Care Alert Management - Decision Support by a Logistic Regression Model

The Elder Care project implies the use of a body sensor and a location device by each elderly. This architecture will originate a very large data set that should be treated by statistical methods. We purpose a logistic model approach, that will be used during the prototype implementation phase and after that in the comercial phase. In the prototype implementation phase (already running), an elders sample is going to be analyzed to eliminate some redundant or useless covariates, allowing a spare in the final equipment complexity and price. Besides, logistic model has an easy odds ratio interpretation, and so we will be able to determine elders problems risk factors. After the prototype implementation phase, we hope to have a better knowledge of elders risk factors and important covariates that must be considered. Again, the logistic model may be applied to correctly decide whenever an alarm signal should be sent for some monitorized elderly, preventing sensibility and specificity issues. The presented decision model will be used by a multi-agent system, in order to make reliable decisions using fewer resources.

Application of Statistical Methods to Improve an Acceleration Based Algorithm

Falls are the leading reason for death related accidents in people over 65 years old. Concerning this situation, it is necessary to develop a viable way of detecting these falls as fast as possible, so that medical assistance can be provided within useful time.

Some results on Gaussian mixtures

We investigate Gaussian mixtures with independent components, whose parameters are numerically estimated. A decomposition of a Gaussian mixture is presented when the components have a common variance. We introduce a shifted and scaled t-Student distribution as an approximation for the distribution of Gaussian mixtures when their components have a common mean and develop a hypothesis test for testing the equality of the components means. Finally, we analyse the fitness of the approximate model to the logarithmic daily returns of the Portuguese stock index PSI-20.

Maximum likelihood estimation in pooled sample tests

Pooled sample tests, firstly used on the classification problem (identifying all individuals with some characteristic), may also be applied to estimate the prevalence rate. Moreover, the pooled sample methods may attain greater efficiency when applied to estimate some prevalence rate, since it is no longer necessary to perform any individual test. We develop a maximum likelihood computational algorithm for the prevalence rate estimation, and we analyze its performance.

Exploring Pareto scale mixtures

Pareto scale mixtures can be used to obtain distributions with heavier tails. An explanation of this model properties is provided, together with a discussion about the parameters estimation. Finally, a real data application is presented, consisting in the larger earthquakes seismic moment modeling.

Improving the Elder Care’s Wireless Sensor Network Fall Detection System Using Logistic Regression

The world’s population is aging; we are already facing many socioeconomic challenges directly related to this problem. These challenges will only tend to grow as time passes. If viable solutions are not found in time, these challenges will become unbearable as the elderly population surpasses the younger population.

Pseudo-convex mixtures generated by shape-extended stable distributions for extremes

A shape-extended definition of min-stable (and max-stable) laws is provided and applied to obtain pseudo-convex mixtures. The main properties of the obtained class of distributions are derived, and a particular case generated by the exponential distribution is investigated.