Indoor Place Classification Using Robot Behavior and Vision Data

The behavioral patterns of human in buildings influence the rational setting of space and function dramatically. However, due to the lack of data acquisition methods and data accuracy, big data analysis and visualization research in the microscopic aspects of indoor space is hampered. With the maturity of indoor positioning technology, UWB (Ultra Wideband) positioning technology based on narrow pulse has the characteristics of high transmission rate, low transmit power and strong penetrating ability, which provides more accurate results for the behavior data acquisition in indoor space. In this research, the big data thinking has been introduced into the behavioral performance analysis process. Therefore, data acquisition, data storage and management, behavioral data visualization and machine learning algorithms are integrated into a set of behavioral data analysis and visualization system, to quantitative research the behavioral characteristics of visitors in the exhibition hall by the on-site experiment .

In this study, we aimed to analyze the in-store behavior of customers at a home improvement retail company using data collected from Bluetooth Low Energy beacon devices installed on shelves and shopping carts within a selected store. The beacons were strategically placed on store shelves to ensure complete coverage, leaving no blind spots. To cover 18 departments spanning a total area of approximately 4,800 square meters, 99 beacons were deployed. The duration of stay in each department, the order of visits, and the absolute visit date and time were recorded in the database. To investigate the relationship between in-store behavior and purchase data, we combined customers' behavioral data with their purchase information. Correlation analysis revealed a positive relationship between visit duration and purchase amount, particularly in the Floor Deco, Paint, and Taps departments. Additionally, we visualized store-wide data using a network diagram, highlighting key shopping areas, customer flow patterns, and high-revenue departments. The problem was also formulated as a multi-class classification task, and LSTM and XGBoost algorithms were applied for comparative analysis. Experiments were conducted on both the original dataset and a cleaned version, utilizing two distinct data modeling approaches: one based solely on sequential department visits and another incorporating visit duration. The results showed that both models performed similarly on the noisy dataset, indicating that adding duration information did not improve learning. However, when trained on the cleaned dataset where short- duration visits were removed, LSTM models outperformed XGBoost, demonstrating a stronger ability to capture meaningful sequential patterns. These findings highlight the potential of BLE beacon technology in retail analytics, offering deeper insights into customer behavior and informing data-driven decision-making for store optimization and personalized marketing. Future work will focus on expanding the dataset and refining predictive models to further enhance the accuracy and applicability of in-store behavior analysis.

Identifying high-risk students as early as possible plays an important role in improving the quality of education. To do this, most of the existing research used traditional machine learning algorithms to predict student achievement based on their behavioral data, from which behavioral features were manually extracted using the experience and knowledge of experts. However, due to the increase in diversity and the overall volume of behavioral data, it is becoming increasingly difficult to identify high-quality handcrafted items. In this article, the authors propose an end-to-end deep learning model that automatically extracts features from heterogeneous student behavior data from multiple sources to predict academic achievement. The key innovation of this model is that it uses long-short-term memory networks to capture the inherent characteristics of the time series for each behavior, and it also uses 2D convolutional networks to extract correlation features between different behaviors. The authors carried out experiments with four types of data on the daily behavior of RTU MIREA students. The experimental results demonstrated that the proposed deep model method outperforms several machine learning algorithms (by about 5 times).

Predicting response to treatment and disease-specific deaths are key tasks in cancer research yet there is a lack of methodologies to achieve these. Large-scale ’omics and digital pathology technologies have led to the need for effective statistical methods for data fusion to extract the most useful patterns from these diverse data types. We present FusionGP, a method for combining heterogeneous data types designed specifically for predicting outcome of treatment and disease. FusionGP is a Gaussian process model that includes a generalization of feature selection for biomarker discovery, allowing for simultaneous, sparse feature selection across multiple data types. Importantly, it can accommodate highly nonlinear structure in the data, and automatically infers the optimal contribution from each input data type. FusionGP compares favourably to several popular classification methods, including the Random Forest classifier, a stepwise logistic regression model and the Support Vector Machine on single data types. By combining gene expression, copy number alteration and digital pathology image data in 119 estrogen receptor (ER)-negative and 345 ER-positive breast tumours, we aim to predict two important clinical outcomes: death and chemoinsensitivity. While gene expression data give the best predictive performance in the majority of cases, the digital pathology data are much better for predicting death in ER cases. Thus, FusionGP is a new tool for selecting informative features from heterogeneous data types and predicting treatment response and prognosis.

In the environment of human-computer interaction of information systems, people are paying more attention to user identity authentication based on operation behaviors. Behavior science research shows that each user has a his/her own behavioral pattern that reflects the unique habits, and maintains stability over a period. As known, most of the previous research have explored the user’s behavior using static authentication models. However, the user’s behavior is evolutionary, even the same user will develop different behavioral tendencies under various times and conditions (job position change or promotion, business content change, increase in age, etc.), causing the difficulty of user authentication under the evolution of user’s behavior. This paper proposes a method named time-lapse detection attempting to establish the authentication model based on the evolution of user’s behavior. We obtained the log data of several years period of the information system of a publishing house. Firstly, we extracted the data of employees’ early operation behaviors and the Bayesian network is used to identify a detection model. Next, the behavior data are divided into multiple test sets according to the time series, and multiple authentication models are carried out to observe the change of authentication accuracy over time. The result shows that, for employees with stable positions and business content, the characteristics of their behavior patterns will change when the number of interactions increases. Moreover, the consequences of the initial detection model fluctuate to different degrees, reducing the accuracy of authentication. Therefore, in future we need to grasp the rules of user behavior and continue to optimize the existing authentication methods of information systems.

All raingauge measurements give only one-point values. Since the practical applications of rain data always deal with areal measures, the knowledge of areal distribution of rain is very important for correct evaluation of the rain volumes. Two types of rainfall input data are generally used: design rainfall hyetographs and long historical rainfall records. The third type of input data has to be recognized: the observed single rainfall event, during which the consequences such as runoff, flooding or pollution loads have been observed and recorded. Good areal characterization of such single event is necessary if the results are to be generalized. In the paper a set of extreme rainfall events observed in Lund during a period of two years is presented. Point and areal intensities are presented for ten extreme events. Storm centered Areal Reduction Factors (AFR-s) calculated for different durations are also presented. Size, shape and movement of the raincells is discussed in the context of the meteorological background observed.

Work-related stress and burnout among physicians are of increasing relevance. The aim of this study was to investigate work-related behavior and experience patterns and predictors of mental health of physicians working in medical practice in Germany. We surveyed a stratified, random sample of 900 physicians from different specialties. The questionnaire included the standardized instruments Work-related Behavior and Experience Pattern (AVEM) and the Short Form-12 Health Survey (SF-12). Only one third of physicians reported high or very high general satisfaction with their job, but 64% would choose to study medicine again. Only 18% of physicians presented a healthy behavior and experience pattern. Almost 40% presented a pattern of reduced motivation to work, 21% were at risk of overexertion, and 22% at risk for burnout. Willingness to study medicine again, fulfilled job expectations, professional years, marital status, and behavior patterns were significant predictors of mental health and accounted for 35.6% of the variance in mental health scores. Job-related perceptions also had a significant effect on burnout. The strong influence of work-related perceptions suggests a need for realistic expectation management in medical education, as well as support in stress management and coping strategies during medical training.

In recent decade, many state-of-the-art algorithms on image classification as well as audio classification have achieved noticeable successes with the development of deep convolutional neural network (CNN). However, most of the works only exploit single type of training data. In this paper, we present a study on classifying bird species by exploiting the combination of both visual (images) and audio (sounds) data using CNN, which has been sparsely treated so far. Specifically, we propose CNN-based multimodal learning models in three types of fusion strategies (early, middle, late) to settle the issues of combining training data cross domains. The advantage of our proposed method lies on the fact that We can utilize CNN not only to extract features from image and audio data (spectrogram) but also to combine the features across modalities. In the experiment, we train and evaluate the network structure on a comprehensive CUB-200-2011 standard data set combing our originally collected audio data set with respect to the data species. We observe that a model which utilizes the combination of both data outperforms models trained with only an either type of data. We also show that transfer learning can significantly increase the classification performance.

GIS based optimal impervious surface map generation using various spatial data for urban nonpoint source management

Recently both game industry professionals and academic researchers have started focusing on player-generated behavioral data as a mean to gather insights on player psychology through datamining. Although some research has already proven solid correlations between in-game behavior and personality, most techniques focus on extracting knowledge from in-game behavior data alone. This paper posits that triangulating exclusively behavioral datasets with established theoretical frameworks serving as hermeneutic grids, may help extracting additional meaning and information. The hermeneutic grid selected for this study is the Reiss Motivation Profiler and it is applied to behavioral data gathered from Minecraft players.

Search is designed primarily for human use, but we still lack efficient user models to evaluate search systems from the user's perspective. The purpose of this research is to identify user behaviour patterns during the search process, and develop a user model to illustrate the relationships between search behaviour and user perception, in order to predict user engagement in a cost-effective manner by using only behaviour data. To achieve this, we will use three datasets collected from previous studies containing both behavioural data from system log files and perception data. Ultimately, this research will yield a cost-efficient predictive model of user engagement that could be used to more effectively evaluate search systems.

Context Golfo Dulce is a wintering ground of the south-eastern Pacific population of humpback whales. Habitat use and spatial preferences of whales during their reproductive season must be addressed to effectively conserve this species. Aims This study aims to determine spatial segregation depending on humpback whale group composition (groups that included mother–calf dyads versus adult-only groups). Methods Spatial aggregation was assessed using sightings per unit of effort from June to October, 2010–2019. The group preferences were analysed as a function of water depth and distance from the coast, and habitat use was characterised using behavioural data. Key results Our SPUE analysis showed important locations used by humpback whales. The mother–calf groups were present in waters <80 m deep, and their main behaviour was resting (nursing and social). Adult-only groups were in the deeper waters, and the main behaviours were social breaching and travelling. Spatial preferences and behaviour data allowed the identification of critical nursing areas near potential mating grounds. Conclusions and implications Our study has highlighted the need for inshore habitat protection, and these habitat-use patterns should be considered when designing effective conservation and management strategies for marine spatial planning.

Prediction of humans’ behaviors during a disaster: The Behavioral Pattern during Disaster Indicator (BPDI)

Twenty-four hour activity budgets and patterns of behavior in captive ocelots ( Leopardus pardalis)

The exponentially increasing volumes of health data of different types is a challenge for reaching robust conclusions. The Health Vision platform overcomes this issue by providing advanced and multi-scaling data analysis and visualisation. The platform functions through user defined workflows, each organised into four distinct layers: 1. Raw Data, 2. Data Analytics, 3. Decision Support and 4. Data visualisation. In the Raw Data Layer, data from medical records, healthcare platforms, smart homes, monitoring sensors, and medical devices, clinical features from Electronic Health Records (EHRs), etc. can be considered, while the Data Analytics Layer provides data analysis services for data mining. The Decision Support Layer facilitates the selection of appropriate visualisation methods taking into consideration the specifications of the raw data and the output of the data analysis. Finally, the Data visualisation Layer provides intelligent visualisation methodology in order for the user to interact with the data and gain meaningful results. Visualisation is supported by a reactive workflow mechanism, allowing the composition of analysis and visualisation services in a custom application-specific data flow. We demonstrate and evaluate this novel approach in several use cases, where different types of data are involved, including: Clustering of Sjogren's syndrome patients, visualisation of Parkinson's disease patients, mutation association analysis in Chronic Lymphocytic Leukemia patients, behavioral analysis of asthma patients and clinical analysis based on clinical similarities of older adults.