Iram Fatima

A Unified Framework for Activity Recognition-Based Behavior Analysis and Action Prediction in Smart Homes

In recent years, activity recognition in smart homes is an active research area due to its applicability in many applications, such as assistive living and healthcare. Besides activity recognition, the information collected from smart homes has great potential for other application domains like lifestyle analysis, security and surveillance, and interaction monitoring. Therefore, discovery of users common behaviors and prediction of future actions from past behaviors become an important step towards allowing an environment to provide personalized service. In this paper, we develop a unified framework for activity recognition-based behavior analysis and action prediction. For this purpose, first we propose kernel fusion method for accurate activity recognition and then identify the significant sequential behaviors of inhabitants from recognized activities of their daily routines. Moreover, behaviors patterns are further utilized to predict the future actions from past activities. To evaluate the proposed framework, we performed experiments on two real datasets. The results show a remarkable improvement of 13.82% in the accuracy on average of recognized activities along with the extraction of significant behavioral patterns and precise activity predictions with 6.76% increase in F-measure. All this collectively help in understanding the users” actions to gain knowledge about their habits and preferences.

Smart CDSS: integration of Social Media and Interaction Engine (SMIE) in healthcare for chronic disease patients

Chronic disease may lead to life threatening health complications like heart disease, stroke, and diabetes that diminish the quality of life. CDSS (Clinical Decision Support System) helps physician in effective utilization of patient’s clinical information at the time of diagnosis and medication. This paper points out the importance of social media and interaction integration in existing Smart CDSS for chronic diseases. The proposed system monitors health conditions, emotions and interests of patients from patients’ tweets, trajectory and email analysis. We extract keywords, concepts and sentiments from patient’s tweets data. Trajectory analysis identifies the focused activities after considering imperative location and semantic tags. Email analysis finds interesting patterns and communication trends from daily routine of patient. All these outputs are supplied to Smart CDSS into vMR (virtual Medical Record) format through social media adapter. This helps the health practitioners to understand the behavior and lifestyle of patients for better decision making about treatment. Consequently, patients can get continuous relevant recommendations from Smart CDSS based on their personalized profile. To verify and validate the working of proposed methodology, we have implemented a proof of concept prototype that reflects its complete working with potential outcomes.

Analysis and effects of smart home dataset characteristics for daily life activity recognition

Over the last few years, activity recognition in the smart home has become an active research area due to the wide range of human centric-applications. With the development of machine learning algorithms for activity classification, dataset is significantly important for algorithms testing and validation. Collection of real data is a challenging process due to involved budget, human resources, and annotation cost that’s why mostly researchers prefer to utilize existing datasets for evaluation purposes. However, openly available smart home datasets indicate variation in terms of performed activities, deployed sensors, and environment settings. Unfortunately, the analysis of existing datasets characteristic is a bottleneck for researchers while selecting datasets of their intent. In this paper, we develop a Framework for Smart Homes Dataset Analysis (FSHDA) to reflect their diverse dimensions in predefined format. It analyzes a list of data dimensions that covers the variations in time, activities, sensors, and inhabitants. For validation, we examine the effects of proposed data dimension on state-of-the-art activity recognition techniques. The results show that dataset dimensions highly affect the classifiers’ individual activity label assignments and their overall performances. The outcome of our study is helpful for upcoming researchers to develop a better understanding about the smart home datasets characteristics with classifier’s performance.

Classifier ensemble optimization for human activity recognition in smart homes

Recognizing human activities is an active research area due to its applicability in many applications, such as assistive living and healthcare. Currently, the major challenges in activity recognition include the reliability of prediction of each classifier as they differ according to smart homes characteristics. It is not possible that one classifier always performs better than all the other classifiers for every possible situation. Therefore, in this paper, a method for activity recognition is proposed by optimizing the output of multiple classifiers with evolutionary algorithm. We combine the measurement level output of different classifiers in terms of weights for each activity class to make up the ensemble. Classifier ensemble learner generates activity rules by optimizing the prediction accuracy of weighted feature vectors to obtain significant improvement over raw classification. For the evaluation of the proposed method, experiments are performed on two real datasets from CASAS smart home. The results show that our method systematically outperforms single classifier and traditional multiclass models.

Weed classification based on Haar wavelet transform via k-nearest neighbor (k-NN) for real-time automatic sprayer control system

Weeds cause harm to crops by competing for water, light, nutrients and space, reducing crop yields and inhibiting the efficiency of machinery. Although there is a large volume of work on this topic, previous studies have lacked accuracy and efficiency. In this paper an algorithm has been developed and analyzed for real-time specific weeds discrimination that is employed Haar Wavelet Transform (HWT). There are three stages of this paper, segmentation stage, feature extraction stage, and classification stage. Based on the proposed algorithm, 200 highest and informative coefficients are extracted at feature extraction stage, then at the classification stage, a well known classifier k-NN (k-Nearest Neighbor, for which k = 2) has been applied. The proposed method was tested on the database of 200 samples of each category. Furthermore, the result of our proposed method improves the performance by 7--10% as compared with some of the existing techniques that used histogram maxima with threshold, and angular cross sectional intensities. In conclusion, the proposed technique is more accurate at weed leaf classification than the existing techniques when images are captured by a CCD camera. The overall accuracy and efficiency utilizing the proposed method using haar wavelet transform is 94% and 40 ms respectively.

MODM: multi-objective diffusion model for dynamic social networks using evolutionary algorithm

A lot of research efforts have been made to model the diffusion process in social networks that varies from adoption of products in marketing strategies to disease and virus spread. Previously, a diffusion process is usually considered as a single-objective optimization problem, in which different heuristics or approximate algorithms are applied to optimize an objective of spreading single piece of information that captures the notion of diffusion. However, in real social networks individuals simultaneously receive several pieces of information during their communication. Single-objective solutions are inadequate for collective spread of several information pieces. Therefore, in this paper, we propose a Multi-Objective Diffusion Model (MODM) that allows the modeling of complex and nonlinear phenomena of multiple types of information exchange, and calculate the information worth of each individual from different aspects of information spread such as score, influence and diversity. We design evolutionary algorithm to achieve the multi-objectives in single diffusion process. Through extensive experiments on a real world data set, we have observed that MODM leads to a richer and more realistic class of diffusion model compared to a single objective. This signifies the correlation between the importance of each individual and his information processing capability. Our results indicate that some individuals in the network are naturally and significantly better connected in terms of receiving information irrespective of the starting position of the diffusion process.

A Sleep Monitoring Application for u-lifecare Using Accelerometer Sensor of Smartphone

Ubiquitous lifecare (u-lifecare) is regarded as a seamless technology that can provide services to the patients as well as facilitate the healthy people to maintain an active lifestyle. In this paper, we develop a sleep monitoring application to assists the healthy people for managing their sleep. It provides an unobtrusive and proactive way for the self-management. We utilize the embedded accelerometer sensor of the smartphone as a client node to collect the sleeping data logs. Our proposed model is server-driven approach and process the data over the server machine. We classify the body movements and compute the useful sleep analytics. It facilitates the users to keep the record of daily sleep and assists to change their unhealthy sleeping habits that are identified by our computed sleep analytics such as bed time, wake up, fell asleep, body movements, frequent body movements at different stages of the night, sleep efficiency and time spent in the bed. Furthermore, we also provide our pilot study results to demonstrate the applicability with the real-world service scenarios.

Activity recognition: an evolutionary ensembles approach

Activity recognition is an emerging field that demands active research in ubiquitous computing for analyzing complex scenarios such as concurrent situation assessment and domination of major over the minor activities. In this paper, an evolutionary ensembles approach using Genetic Algorithm (GA) as a homogeneous learner has been proposed. This approach values both minor and major activities by processing each of them independently. It consists of two phases. The first phase is preprocessing of sensory data and extraction of feature vectors. Evolutionary ensembles are designed in second phase to learn different daily life activities. Finally, multiple ensembles output is pooled on central node as a complete rule profile for all performed activities. The proposed approach was evaluated on six different types of activities from Intelligent System Laboratory (ISL) dataset. It shows a higher accuracy as compared to single learner GA.

Activity Recognition Based on SVM Kernel Fusion in Smart Home

Smart home is regarded as an independent healthy living for elderly person and it demands active research in activity recognition. This paper proposes kernel fusion method, using Support Vector Machine (SVM) in order to improve the accuracy of performed activities. Although, SVM is a powerful statistical technique, but still suffer from the expected level of accuracy due to complex feature space. Designing a new kernel function is difficult task, while common available kernel functions are not adequate for the activity recognition domain to achieve high accuracy. We introduce a method, to train the different SVMs independently over the standard kernel functions and fuse the individual results on the decision level to increase the confidence of each activity class. The proposed approach has been evaluated on ten different kinds of activities from CASAS smart home (Tulum 2009) real world dataset. We compare our SVM kernel fusion approach with the standard kernel functions and get overall accuracy of 91.41 %.

Effects of Smart Home Dataset Characteristics on Classifiers Performance for Human Activity Recognition

Over the last few years, activity recognition in the smart home has become an active research area due to the wide range of human centric-applications. A list of machine learning algorithms is available for activity classification. Datasets collected in smart homes poses unique challenges to these methods for classification because of their high dimensionality, multi-class activities and various deployed sensors. In fact the nature of dataset plays considerable role in recognizing the activities accurately for a particular classifier. In this paper, we evaluated the effects of smart home datasets characteristics on state-of-the-art activity recognition techniques. We applied probabilistic and statistical methods such as the Artificial Neural network, Hidden Markov Model, Conditional Random Field, and Support Vector Machines. The four real world datasets are selected from three most recent and publically available smart home projects. Our experimental results show that how the performance of activity classifiers are influenced by the dataset characteristics. The outcome of our study will be helpful for upcoming researchers to develop a better understanding about the smart home datasets characteristics in combination with classifier’s performance.