Location-based activity behavior deviation detection for nursing home using IoT devices

Question

1. What is the main objective of the system designed to monitor the deviated activity behavior of nursing home residents?

2. What are the four common types of monitoring techniques?

3. What are common methodologies in studying deviated activity behavior?

4. What components are included in the proposed hardware setup?

Accepted Answer

The main objective of the system designed to monitor the deviated activity behavior of nursing home residents is to identify the residents' activity behavior and determine the irregular activity behaviors. The system aims to differentiate their activity between private and group activity when computing their deviated activity behavior. It also addresses the constraints of monitoring residents' activity behavior in a nursing home, such as the building structure and the influence of group activities or community. The system does not have ground truth on the data collected, resulting in the accuracy of the system output not being validated. Additionally, the identity of the nursing home residents is anonymized to comply with the Singapore Personal Data Protection Act. Therefore, unsupervised knowledge extraction is more desired when compared to the supervised knowledge extraction model. The system uses a building-scale monitoring system, wearable card tags with Bluetooth beacons, and a data fusion method to generate the hybrid norm based on individual and group norms. This allows for the extraction and analysis of deviated activity behavior, which does not follow the normal daily pattern of a resident. The system's key contributions include studying the resident's activity behavior in a nursing home, proposing a data fusion method to identify the daily norm, and performing empirical analysis on the deviated activity behavior using rules-based classification.

Accepted Answer

The four common types of monitoring techniques are people-driven, event-driven, location-driven, and data-driven. People-driven techniques use humans as the main source of generating information, often involving sensors in wearable devices. Event-driven techniques focus on daily life activities and use models like HMM and LSTM to detect deviations. Location-driven techniques involve installing sensors in specific areas, such as motion sensors and RFID sensors. Data-driven techniques combine various information sources, including physical sensors and cyber data sources, to infer human behavior.

Accepted Answer

Common methodologies in studying deviated activity behavior include prediction model, state estimation model, and clustering and exploration model. Prediction models use statistics to predict potential activity behavior and label deviated behavior if predictions do not match outcomes. Recent methods like LSTM and other statistical models are used for prediction. State estimation models map user behavior into system states, requiring human intervention for mapping. Examples include SOM and HMM models. Clustering and exploration models use steps to extract insights, often used without ground-truth. Examples include hybrid k-means clustering and isolation forest. Each method has its drawbacks, such as the need for large data in prediction models and domain knowledge in clustering models.

Accepted Answer

The proposed hardware setup comprises of card tags with BLE beacons and beacon receivers. The Bluetooth beacon card model, shown in Fig. 1a, is capable of transmitting beacon every 1000ms using Nordic nRF52 chip with a range of up to 40 meters. Each room has a beacon receiver installed to pick up the Bluetooth beacon transmitted. A unique identifier is assigned to each card tag to filter out irrelevant Bluetooth devices. The building consists of 4 levels, with the top 2 levels serving as the residential area and the lower 2 levels as the staff area and basement level 1. Over 138 Bluetooth beacon receivers are installed to monitor residents' activity behavior. A total of 50 residents within the nursing home participate in the study. The beacon receiver uses RedBear Duo to transmit the collected Bluetooth beacon list to the local server for further processing.

Accepted Answer

The preprocessing steps in the data processing architecture include time-sync, location encoding, windows fitting, and data smoothing. Time-sync synchronizes time for data entry, location encoding represents locations numerically, windows fitting fits location data into five-minute windows, and data smoothing removes short duration stay locations. These steps result in a structured residents' trajectory data for further analysis.

Accepted Answer

Based on the study, the ideal number of clusters for nursing home residents is 5. This conclusion was reached by applying the sum of squared distance (SSD) for different number of k values. The SSD measures the similarity between residents in the same cluster, with lower values indicating higher similarity. The study aimed to find clusters within the range of 2 to 7 clusters out of 50 residents. The SSD result, presented in Figure 4, showed that k=5 had the lowest SSD value compared to other choices of k. Using k=5, the clustering algorithm was formulated as shown in Algorithm 3, and the clustering result was presented in Figure 5. The clusters were separated by building level and had their own characteristic. For example, level 3 residents were divided into three different groups, while level 2 residents were divided into two groups. The study also observed differences in residents' activities within each cluster, which will be further analyzed in the next section.

Accepted Answer

To generate hybrid norms, individual norms and group norms are used as data inputs. The individual norm represents the regular pattern of a resident over the data collection period, while the group norm denotes the regular pattern of all clustered residents. The transition method between individual and group norms is addressed by considering the starting and ending periods of group activities. The hybrid norm is computed using Algorithm 4 with a computational complexity of O(n^2).

Accepted Answer

To identify residents' deviated activity behavior, we first compute the adaptive hybrid norm for each user. Then, we extract deviated locations using a filtering function based on the norm and daily input data. The binary comparison method is adopted to compare the input data with the generated adaptive hybrid norm. Algorithm 5 is proposed to compute daily trajectory by iterating the filter function every day. Deviated activity behaviors are highlighted in Fig. 7, showing different locations and times where residents deviate from their hybrid norm. Three different deviated activity behaviors are detected for resident u 21, including visiting public areas, staying in the origin, and going to private areas. By fusing group norm and individual norm, deviated activity behaviors can be computed for every resident in the nursing home.

Accepted Answer

The three different deviated activity behaviors identified in the nursing home's residents are (1) sleep irregularity, (2) awake irregularity, and (3) private visiting. Sleep irregularity denotes the resident going to another location instead of staying back at their origin during nighttime. Awake irregularity denotes the resident either leaving the room late or early compared to their routine schedule. Private visiting denotes the resident going to another staying room instead of the common places cluster or presenting in their usual location.

Accepted Answer

The proposed system differentiates deviated activity behavior by understanding residents' normal activity behavior and extracting deviations. It identifies three categories of deviated activity behavior: sleep irregularity, awake irregularity, and private visiting. By analyzing these deviations, the system can study and detect abnormal behavior patterns. Future research aims to incorporate more data sources for accurate detection and develop a real-time system for dynamic group activity formation. Additionally, improving ground-truth data collection and studying compatibility with different countries' data protection rules are areas of interest for enhancing the deviated activity behavior detection module.

Location-based activity behavior deviation detection for nursing home using IoT devices

Chat with Paper

AI Agents for this Paper

Most frequently asked questions

1. What is the main objective of the system designed to monitor the deviated activity behavior of nursing home residents?

2. What are the four common types of monitoring techniques?

3. What are common methodologies in studying deviated activity behavior?

4. What components are included in the proposed hardware setup?

5. What are the preprocessing steps in the data processing architecture?

6. What is the ideal number of clusters for nursing home residents?

7. How to generate hybrid norms?

8. How to identify residents' deviated activity behavior?

9. What are the three different deviated activity behaviors identified in the nursing home's residents?

10. How does the proposed system differentiate deviated activity behavior?

Citations

Revenue forecasting in smart retail based on customer clustering analysis

Bi-functional glucose Sensing-Transmission (Sens-Tra) sensor for IoT-based applications

References

Toward memory-based reasoning

Predicting ADL disability in community-dwelling elderly people using physical frailty indicators: a systematic review

Similarity measures for categorical data: A comparative evaluation

Applications of Data Mining in Computer Security

A survey of data fusion in smart city applications

Related Papers (5)

Intelligent and Scalable IoT Edge-Cloud System

SODA: A software-defined security framework for IoT environments

FS-IIoTSim: a flexible and scalable simulation framework for performance evaluation of industrial Internet of things systems

An Environment Monitoring System Design Based on IOT and Web 2.0

SUTSEC: SDN Utilized trust based secure clustering in IoT