One of the most common functions smart environments is to monitor and assist older adults with their activities daily living. Activity recognition a key component in this application. It essentially temporal classification problem which has been modeled past by naïve Bayes classifiers hidden Markov models (HMMs). In paper, we describe use another probabilistic model: Conditional Random Fields (CRFs), currently gaining popularity for its remarkable performance activity recognition. Our focus...

An important problem that arises during the data mining process in many new emerging application domains is with temporal dependencies. One such domain activity discovery and recognition. Activity recognition used real world systems, as assisted living security it has been vastly studied recent years. However, features relations which provide useful insights for models have not exploited to their full potential by algorithms. In this paper, we propose a model discovering of patterns from...

Residential load scheduling has been introduced in recent years as a result of implementing demand response programmes the residential sector. Up to now, studies have focussed on developing algorithms for loads with objective reducing electricity payment while satisfying users’ comfort level. To aim at this goal, user should determine desired window each appliance initiate process. Although these designed benefit end‐users, determining beginning process can be challenging. In study, an...

In this paper we develop a reliable system for smart irrigation of greenhouses using artificial neural networks, and an IoT architecture. Our solution uses four sensors in different layers soil to predict future moisture. Using dataset collected by running experiments on soils, show high performance networks compared existing alternative method support vector regression. To reduce the processing power network edge devices, propose transfer learning. Transfer learning also speeds up training...

This paper presents a sequence-based activity prediction approach which uses Bayesian networks in novel two-step process to predict both activities and their corresponding features. In addition the proposed model, we also present results of several search score (S&S) constraint-based (CB) structure learning algorithms. The performance model is compared with naïve Bayes other aforementionedS&S CB experimental are performed on real data collected from smart home over period five months....

The correct recognition of Motor Imagery task in Brain-Computer Interface (BCI) systems has been an important issue recent studies. In this study, we propose a classification framework based on ensemble methods to handle spectral and spatial EEG signal characteristics. A mixture two classifiers used for combining multiple information sources. performance the proposed classifier evaluated two-class problem (right left hand) from BCI Competition IV dataset 2a. features training data are...

Activity recognition from sensor data deals with various challenges, such as overlapping activities, activity labeling, and detection. Although each challenge in the field of has great importance, most important one refers to online recognition. The present study tries use hierarchical hidden Markov model detect an on stream which can predict environment any event. samples were labeled by statistical features duration activity. results our proposed method test two different datasets smart...

Abstract Independent living for elderly people is often viewed as an impossible task, due to the great many perils and difficulties. With advancements of Ambient Intelligence, this scenario no longer out reach smart homes offer a computationally inexpensive solution problem. In paper we address these difficulties propose novel method Anomaly Detection in elderly’s daily routine behavior. Our proposed model ConvLSTM Autoencoder processing spatiotemporal data, given fact that type behavior...

In recent years, learning embeddings for nodes of a graph has become one the most efficient w ays t o solve different problems such as link prediction, clustering and classification. I n his p aper, e ropose ovel m ethod, called SECI, nodes, with application to prediction. SECI samples from network using breadth-first search depth-first s earch, nd i nterpolates b etween these two centrality indices. The intuition behind is that have low score only very small neighborhood explored; dominant...

Video violence recognition based on deep learning concerns accurate yet scalable human recognition. Currently, most state-of-the-art video studies use CNN-based models to represent and categorize videos. However, recent suggest that pre-trained transformers are more than various analysis benchmarks. Yet these not thoroughly evaluated for This paper introduces a novel transformer-based Mixture of Experts (MoE) system. Through an intelligent combination large vision efficient transformer...

This paper proposes an approach to the problem of building block extraction in context evolutionary algorithms (with binary strings). The method is based upon construction a GMDH neural network model population promising solutions with aim extracting blocks from resultant network. operation proposed regardless order by which are positioned strings representing solutions. experiments carried out on some well-known benchmark functions including DeJong's

Fuzzy c-means (FCM) is one of the most popular fuzzy clustering methods and it used in various applications computer science. Most including FCM, suffer from bad initialization problem. If initial cluster centers (membership degree FCM) are not selected appropriately, may yield poor results. In this paper we propose a method called MinMax FCM to overcome A new objective function designed aim. We use maximum variance clusters as function. regard, high-variance penalized. compare with terms...

With the expansion of applications artificial intelligence and machine learning in various areas, many challenges have arisen training algorithms. One most important is imbalanced datasets. The data generally refers to a classification problem where number samples per class not equally distributed. Typically there large amount for one (referred as majority class) much fewer other minority class). In such datasets, algorithms are biased toward reach better accuracy, which leads lack data....

Video violence recognition based on deep learning concerns accurate yet scalable human recognition. Currently, most state-of-the-art video studies use CNN-based models to represent and categorize videos. However, recent suggest that pre-trained transformers are more than various analysis benchmarks. Yet these not thoroughly evaluated for This paper introduces a novel transformer-based Mixture of Experts (MoE) system. Through an intelligent combination large vision efficient transformer...