Combination Of Classifiers Research Articles

Artificial Intelligence-Based Diabetes Diagnosis with Belief Functions Theory

We compared various machine learning (ML) methods, such as the K-nearest neighbor (KNN), support vector machine (SVM), and decision tree and deep learning (DL) methods, like the recurrent neural network, convolutional neural network, long short-term memory (LSTM), and gated recurrent unit (GRU), to determine the ones with the highest precision. These algorithms learn from data and are subject to different imprecisions and uncertainties. The uncertainty arises from the bad reading of data and/or inaccurate sensor acquisition. We studied how these methods may be combined in a fusion classifier to improve their performance. The Dempster–Shafer method, which uses the formalism of belief functions characterized by asymmetry to model nonprecise and uncertain data, is used for classifier fusion. Diagnosis in the medical field is an important step for the early detection of diseases. In this study, the fusion classifiers were used to diagnose diabetes with the required accuracy. The results demonstrated that the fusion classifiers outperformed the individual classifiers as well as those obtained in the literature. The combined LSTM and GRU fusion classifiers achieved the highest accuracy rate of 98%.

IntramuscularEMGclassifier for detecting myopathy and neuropathy

AbstractThis article presents an automatic diagnostic system to classify intramuscular electromyography (iEMG) signals, thereby detecting neuromuscular disorders. To this end, we tailored the center symmetric local binary pattern (CSLBP) to analyze one‐dimensional (‐D) signals. In this approach, the ‐D CSLBP feature extracted from a decimated iEMG signal is fed to a combination of classifiers, which in turn assigns a set of labels to the signal, and ultimately the signal category is determined by the Boyer‐Moore majority voting (BMMV) algorithm. The proposed framework was investigated with a benchmark iEMG dataset that contains signals recorded from three different muscles:biceps brachii(BB),deltoideus(DE), andvastus medialis(VM). In a repeated ‐fold cross‐validation, CSLBP‐Combined‐Classifiers‐BMMV (CSLBP‐CC‐BMMV) achieved an average classification accuracy of %, %, and % for the iEMG signals recorded from BB, DE, and VM muscle, respectively. Interestingly, the performance of CSLBP‐CC‐BMMV surpassed the other published approaches and ensemble learning methods that are akin to our scheme in terms of classification accuracy and computational time.

Selection of Classifiers to Enhance Efficacy of Metal/Organic Hybrid Sensor Array for VOC and Toxic Gas Identification

Modern developments in gas sensor technology include a decrease in size and an increase in sensitivity and selectivity. These improvements, paired with postprocessing tools, such as machine learning, are pushing gas detection toward viability for complex tasks, such as volatile organic compound (VOC) analysis in human breath. In our research, we use a sensor array fabricated in our lab featuring a hybrid combination of metals and organic polymers [palladium (Pd), zinc oxide (ZnO), polypyrrole (PPy), and polyaniline (PANI)] designed to detect a range of VOCs and toxic gases (CO, H2, CH3OH, and NO2). An exhaustive analysis of 25 machine learning classifiers using three different feature sets was completed to find the best classifier and feature set combinations for one versus rest gas classification. We determined that ensemble classifiers, using normalized sensor data as a feature set, yield the best classification results. From these results, we demonstrated that Pd, PPy, and PANI are best suited to identify H2, NO2, and CH3OH, respectively. Furthermore, PANI is best suited to identify CO, so we correctly identified four gases from three sensor materials with sensitivity values all above 85%. These promising classification results could allow us to expand our set of gases and, therefore, make this sensor array viable for real-world applications.

Gait phases recognition based on lower limb sEMG signals using LDA-PSO-LSTM algorithm

Gait phases are widely used in exoskeleton movement control. Surface electromyography (sEMG) is predictive and plays an important role in gait phase recognition. The purpose of this study is to improve the stability and accuracy of gait recognition methods based on the sEMG signals of lower limbs. First, we presented a LDA-PSO-LSTM algorithm based on feature combination selection and verified its recognition accuracy through experiments. LDA-PSO-LSTM had an average recognition rate of94.89% and a maximum accuracy of 97.02%. Second, we tested and compared the recognition accuracy of LDA-LSTM (92.17%). Experiments showed that the PSO optimization model had good recognition performance. Finally, we compared LDA-LSTM with all classifier combinations and concluded that the LDA-LSTM method has the highest recognition rate among a series of method combinations. The results indicated that LDA-PSO-LSTM as a classification model has apparent advantages in gait recognition. LDA-PSO-LSTM provides more accurate gait phase results for lower limb exoskeleton control. This method is beneficial to the development of the exoskeleton gait recognition system.

Collaboration graph for feature set partitioning in data classification

The curse of dimensionality of features in data classification is still an open issue. An approach to solve this problem is to partition features into several sub-sets of features hence the data classification task for every subset is performed. Then, an ensemble of these classifications are reported as the result of the classification problem. However, the feature set partitioning into sub-sets of features is still an area of research interest. Thus, in this paper, an innovative framework is proposed in which, first, a collaboration measure between each two features is defined and measured. Then, the collaboration graph, consisted of features as nodes and measured collaborations as edges’ weights, is generated according to the collaboration measures calculated. After that, a community detection method is used to find the graph communities. The communities are considered as the feature subsets and a base classifier is trained for each subset based on the corresponding training data of the subsets. Then, the ensemble classifier is created by a combination of base classifiers according to the AdaBoost Aggreagation. The simulation results of the proposed approach over the real and synthetic datasets indicate that the proposed approach considerably increases the classification accuracy in comparison to previous methods.

Cyberbullying Identification System Based Deep Learning Algorithms

Cyberbullying is characterized by deliberate and sustained peer aggression, as well as a power differential between the victim and the perpetrators or abusers. Cyberbullying can have a variety of consequences for victims, including mental health problems, poor academic performance, a tendency to drop out of work, and even suicidal thoughts. The main objective of this study was to develop a cyberbullying detection system (CDS) to uncover hateful and abusive behaviour on social media platforms. Two experiments were carried out to train and test the proposed system with binary and multiclass cyberbullying classification datasets. Hybrid deep learning architecture consisting of convolutional neural networks integrated with bidirectional long short-term memory networks (CNN-BiLSTM) and single BiLSTM models were compared in terms of their ability to classify social media posts into several bullying types related to gender, religion, ethnicity, age, aggression, and non-cyberbullying. Both classifiers showed promising performance in the binary classification dataset (aggressive or non-aggressive bullying), with a detection accuracy of 94%. For the multiclass dataset, BiLSTM outperformed the combined CNN-BiLSTM classifier, achieving an accuracy of 99%. A comparison of our method to the existing method on the multiclass classification dataset revealed that our method performed better in detecting online bullying.

Controlling Perceived Emotion in Symbolic Music Generation with Monte Carlo Tree Search

This paper presents a new approach for controlling emotion in symbolic music generation with Monte Carlo Tree Search. We use Monte Carlo Tree Search as a decoding mechanism to steer the probability distribution learned by a language model towards a given emotion. At every step of the decoding process, we use Predictor Upper Confidence for Trees (PUCT) to search for sequences that maximize the average values of emotion and quality as given by an emotion classifier and a discriminator, respectively. We use a language model as PUCT's policy and a combination of the emotion classifier and the discriminator as its value function. To decode the next token in a piece of music, we sample from the distribution of node visits created during the search. We evaluate the quality of the generated samples with respect to human-composed pieces using a set of objective metrics computed directly from the generated samples. We also perform a user study to evaluate how human subjects perceive the generated samples' quality and emotion. We compare PUCT against Stochastic Bi-Objective Beam Search (SBBS) and Conditional Sampling (CS). Results suggest that PUCT outperforms SBBS and CS in almost all metrics of music quality and emotion.

An Efficient Method Combined Data-Driven for Detecting Electricity Theft with Stacking Structure Based on Grey Relation Analysis

Nowadays, electricity theft has been a major problem worldwide. Although many single-classification algorithms or an ensemble of single learners (i.e., homogeneous ensemble learning) have proven able to automatically identify suspicious customers in recent years, after the accuracy of these methods reaches a certain level, it still cannot be improved even if it continues to be optimized. To break through this bottleneck, a heterogeneous ensemble learning method with stacking integrated structure of different strong individual learners for detection of electricity theft is presented in this paper. Firstly, we use the grey relation analysis (GRA) method to select the heterogeneous strong classifier combination of LG + LSTM + KNN as the base model layer of stacking structure based on the principle of the highest comprehensive evaluation index value. Secondly, the support vector machine (SVM) model with relatively good results of the stacking overall structure experiment is selected as the model of the meta-model layer. In this way, a heterogeneous integrated learning model for electricity theft detection of the stacking structure is constructed. Finally, the experiments of this model are conducted on electricity consumption data from State Grid Corporation of China, and the results show that the detection performance of the proposed method is better than that of the existing state-of-the-art detection method (where the area under receiver operating characteristic curve (AUC) value is 0.98675).

A hierarchical machine learning model based on Glioblastoma patients' clinical, biomedical, and image data to analyze their treatment plans

Aim of studyGlioblastoma Multiforme (GBM) is an aggressive brain cancer in adults that kills most patients in the first year due to ineffective treatment. Different clinical, biomedical, and image data features are needed to analyze GBM, increasing complexities. Besides, they lead to weak performances for machine learning models due to ignoring physicians' knowledge. Therefore, this paper proposes a hierarchical model based on Fuzzy C-mean (FCM) clustering, Wrapper feature selection, and twelve classifiers to analyze treatment plans. Methodology/ApproachThe proposed method finds the effectiveness of previous and current treatment plans, hierarchically determining the best decision for future treatment plans for GBM patients using clinical data, biomedical data, and different image data. A case study is presented based on the Cancer Genome Atlas Glioblastoma Multiforme dataset to prove the effectiveness of the proposed model. This dataset is analyzed using data preprocessing, experts' knowledge, and a feature reduction method based on the Principal Component Analysis. Then, the FCM clustering method is utilized to reinforce classifier learning. Outcomes of studyThe proposed model finds the best combination of Wrapper feature selection and classifier for each cluster based on different measures, including accuracy, sensitivity, specificity, precision, F-score, and G-mean according to a hierarchical structure. It has the best performance among other reinforced classifiers. Besides, this model is compatible with real-world medical processes for GBM patients based on clinical, biomedical, and image data.

Class-Imbalanced Deep Learning via a Class-Balanced Ensemble.

Class imbalance is a prevalent phenomenon in various real-world applications and it presents significant challenges to model learning, including deep learning. In this work, we embed ensemble learning into the deep convolutional neural networks (CNNs) to tackle the class-imbalanced learning problem. An ensemble of auxiliary classifiers branching out from various hidden layers of a CNN is trained together with the CNN in an end-to-end manner. To that end, we designed a new loss function that can rectify the bias toward the majority classes by forcing the CNN's hidden layers and its associated auxiliary classifiers to focus on the samples that have been misclassified by previous layers, thus enabling subsequent layers to develop diverse behavior and fix the errors of previous layers in a batch-wise manner. A unique feature of the new method is that the ensemble of auxiliary classifiers can work together with the main CNN to form a more powerful combined classifier, or can be removed after finished training the CNN and thus only acting the role of assisting class imbalance learning of the CNN to enhance the neural network's capability in dealing with class-imbalanced data. Comprehensive experiments are conducted on four benchmark data sets of increasing complexity (CIFAR-10, CIFAR-100, iNaturalist, and CelebA) and the results demonstrate significant performance improvements over the state-of-the-art deep imbalance learning methods.

Comparative study of term-weighting schemes for environmental big data using machine learning

Widely-used term-weighting schemes and machine learning (ML) classifiers with default parameter settings were assessed for their performance when applied to environmental big data analysis. Five term-weighting schemes [term frequency (TF), TF–inverse document frequency (TF-IDF), Best Match 25 (BM25), TF–inverse gravity moment (TF-IGM), and TF–IDF–inverse class frequency (TF-IDF-ICF)] and five different ML classifiers [support vector machine (SVM), Naive Bayes (NB), logistic regression (LR), random forest (RF), and extreme gradient boosting (XGBoost)] were tested. The optimal text-classification scheme and classifier were TF-IDF-ICF and LR, respectively. Based on evaluation criteria, their combination resulted in the best performance of all scheme and classifier combinations for the full environmental data analysis. Category classification performance differed according to the environmental section (climate, air, water, or waste/garbage), with the best performance being achieved for climate, and the poorest for water. This demonstrated the importance of selecting term-weighting schemes and ML classifiers in human-generated environmental big data analysis.

Automatic Detection of Cracks in Cracked Tooth Based on Binary Classification Convolutional Neural Networks.

Cracked tooth syndrome is a commonly encountered disease in dentistry, which is often accompanied by dramatic painful responses from occlusion and temperature stimulation. Current clinical diagnostic trials include traditional methods (such as occlusion test, probing, cold stimulation, etc.) and X-rays based medical imaging (periapical radiography (PR), cone-beam computed tomography (CBCT), etc.). However, these methods are strongly dependent on the experience of the clinicians, and some inconspicuous cracks are also extremely easy to be overlooked by visual observation, which will definitely affect the subsequent treatments. Inspired by the achievements of applying deep convolutional neural networks (CNNs) in crack detection in engineering, this article proposes an image-based crack detection method using a deep CNN classifier in combination with a sliding window algorithm. A CNN model is designed by modifying the size of the input layer and adding a fully connected layer with 2 units based on the ResNet50, and then, the proposed CNN is trained and validated with a self-prepared cracked tooth dataset including 20,000 images. By comparing validation accuracy under seven different learning rates, 10−5 is chosen as the best learning rate for the following testing process. The trained CNN is tested on 100 images with 1920 × 1080-pixel resolutions, which achieves an average accuracy of 90.39%. The results show that the proposed method can effectively detect cracks in images under various conditions (stained, overexplosion, images affected by other diseases). The proposed method in this article provides doctors with a more intelligent diagnostic solution, and it is not only suitable for optical photographs but also for automated diagnosis of other medical imaging images.

Combining autophagy and immune characterizations to predict prognosis and therapeutic response in lung adenocarcinoma.

BackgroundAutophagy, a key regulator of programmed cell death, is critical for maintaining the stability of the intracellular environment. Increasing evidence has revealed the clinical importance of interactions between autophagy and immune status in lung adenocarcinoma. The present study evaluated the potential of autophagy-immune-derived biomarkers to predict prognosis and therapeutic response in patients with lung adenocarcinoma.MethodsPatients from the GSE72094 dataset were randomized 7:3 to a training set and an internal validation set. Three independent cohorts, TCGA, GSE31210, and GSE37745, were used for external verification. Unsupervised hierarchical clustering based on autophagy- and immune-associated genes was used to identify autophagy- and immune-associated molecular patterns, respectively. Significantly prognostic autophagy-immune genes were identified by LASSO analysis and by univariate and multivariate Cox regression analyses. Differences in tumor immune microenvironments, functional pathways, and potential therapeutic responses were investigated to differentiate high-risk and low-risk groups.ResultsHigh autophagy status and high immune status were associated with improved overall survival. Autophagy and immune subtypes were merged into a two-dimensional index to characterize the combined prognostic classifier, with 535 genes defined as autophagy-immune-related differentially expressed genes (DEGs). Four genes (C4BPA, CD300LG, CD96, and S100P) were identified to construct an autophagy-immune-related prognostic risk model. Survival and receiver operating characteristic (ROC) curve analyses showed that this model was significantly prognostic of survival. Patterns of autophagy and immune genes differed in low- and high-risk patients. Enrichment of most immune infiltrating cells was greater, and the expression of crucial immune checkpoint molecules was higher, in the low-risk group. TIDE and immunotherapy clinical cohort analysis predicted that the low-risk group had more potential responders to immunotherapy. GO, KEGG, and GSEA function analysis identified immune- and autophagy-related pathways. Autophagy inducers were observed in patients in the low-risk group, whereas the high-risk group was sensitive to autophagy inhibitors. The expression of the four genes was assessed in clinical specimens and cell lines.ConclusionsThe autophagy-immune-based gene signature represents a promising tool for risk stratification in patients with lung adenocarcinoma, guiding individualized targeted therapy or immunotherapy.

Evaluation of Disability Progression in Multiple Sclerosis via Magnetic-Resonance-Based Deep Learning Techniques.

Short-term disability progression was predicted from a baseline evaluation in patients with multiple sclerosis (MS) using their three-dimensional T1-weighted (3DT1) magnetic resonance images (MRI). One-hundred-and-eighty-one subjects diagnosed with MS underwent 3T-MRI and were followed up for two to six years at two sites, with disability progression defined according to the expanded-disability-status-scale (EDSS) increment at the follow-up. The patients’ 3DT1 images were bias-corrected, brain-extracted, registered onto MNI space, and divided into slices along coronal, sagittal, and axial projections. Deep learning image classification models were applied on slices and devised as ResNet50 fine-tuned adaptations at first on a large independent dataset and secondly on the study sample. The final classifiers’ performance was evaluated via the area under the curve (AUC) of the false versus true positive diagram. Each model was also tested against its null model, obtained by reshuffling patients’ labels in the training set. Informative areas were found by intersecting slices corresponding to models fulfilling the disability progression prediction criteria. At follow-up, 34% of patients had disability progression. Five coronal and five sagittal slices had one classifier surviving the AUC evaluation and null test and predicted disability progression (AUC > 0.72 and AUC > 0.81, respectively). Likewise, fifteen combinations of classifiers and axial slices predicted disability progression in patients (AUC > 0.69). Informative areas were the frontal areas, mainly within the grey matter. Briefly, 3DT1 images may give hints on disability progression in MS patients, exploiting the information hidden in the MRI of specific areas of the brain.

Age-dependent white matter microstructural disintegrity in autism spectrum disorder.

There has been increasing evidence of White Matter (WM) microstructural disintegrity and connectome disruption in Autism Spectrum Disorder (ASD). We evaluated the effects of age on WM microstructure by examining Diffusion Tensor Imaging (DTI) metrics and connectome Edge Density (ED) in a large dataset of ASD and control patients from different age cohorts. N = 583 subjects from four studies from the National Database of Autism Research were included, representing four different age groups: (1) A Longitudinal MRI Study of Infants at Risk of Autism [infants, median age: 7 (interquartile range 1) months, n = 155], (2) Biomarkers of Autism at 12 months [toddlers, 32 (11)m, n = 102], (3) Multimodal Developmental Neurogenetics of Females with ASD [adolescents, 13.1 (5.3) years, n = 230], (4) Atypical Late Neurodevelopment in Autism [young adults, 19.1 (10.7)y, n = 96]. For each subject, we created Fractional Anisotropy (FA), Mean- (MD), Radial- (RD), and Axial Diffusivity (AD) maps as well as ED maps. We performed voxel-wise and tract-based analyses to assess the effects of age, ASD diagnosis and sex on DTI metrics and connectome ED. We also optimized, trained, tested, and validated different combinations of machine learning classifiers and dimensionality reduction algorithms for prediction of ASD diagnoses based on tract-based DTI and ED metrics. There is an age-dependent increase in FA and a decline in MD and RD across WM tracts in all four age cohorts, as well as an ED increase in toddlers and adolescents. After correction for age and sex, we found an ASD-related decrease in FA and ED only in adolescents and young adults, but not in infants or toddlers. While DTI abnormalities were mostly limited to the corpus callosum, connectomes showed a more widespread ASD-related decrease in ED. Finally, the best performing machine-leaning classification model achieved an area under the receiver operating curve of 0.70 in an independent validation cohort. Our results suggest that ASD-related WM microstructural disintegrity becomes evident in adolescents and young adults—but not in infants and toddlers. The ASD-related decrease in ED demonstrates a more widespread involvement of the connectome than DTI metrics, with the most striking differences being localized in the corpus callosum.

Attention emotion recognition via ECG signals

BackgroundPhysiological signal‐based research has been a hot topic in affective computing. Previous works mainly focus on some strong, short‐lived emotions ( e.g., joy, anger), while the attention, which is a weak and long‐lasting emotion, receives less attraction. In this paper, we present a study of attention recognition based on electrocardiogram (ECG) signals, which contain a wealth of information related to emotions.MethodsThe ECG dataset is derived from 10 subjects and specialized for attention detection. To relieve the impact of noise of baseline wondering and power‐line interference, we apply wavelet threshold denoising as preprocessing and extract rich features by pan‐tompkins and wavelet decomposition algorithms. To improve the generalized ability, we tested the performance of a variety of combinations of different feature selection algorithms and classifiers.ResultsExperiments show that the combination of generic algorithm and random forest achieve the highest correct classification rate (CCR) of 86.3%.ConclusionThis study indicates the feasibility and bright future of ECG‐based attention research.

Prediction of Histologic Neoadjuvant Chemotherapy Response in Osteosarcoma Using Pretherapeutic MRI Radiomics.

Histologic response to chemotherapy for osteosarcoma is one of the most important prognostic factors for survival, but assessment occurs after surgery. Although tumor imaging is used for surgical planning and follow-up, it lacks predictive value. Therefore, a radiomics model was developed to predict the response to neoadjuvant chemotherapy based on pretreatment T1-weighted contrast-enhanced MRI. A total of 176 patients (median age, 20 years [range, 5-71 years]; 107 male patients) with osteosarcoma treated with neoadjuvant chemotherapy and surgery between January 2007 and December 2018 in three different centers in France (Centre Léon Bérard in Lyon, Centre Hospitalier Universitaire de Nantes in Nantes, and Hôpital Cochin in Paris) were retrospectively analyzed. Various models were trained from different configurations of the data sets. Two different methods of feature selection were tested with and without ComBat harmonization (ReliefF and t test) to select the most relevant features, and two different classifiers were used to build the models (an artificial neural network and a support vector machine). Sixteen radiomics models were built using the different combinations of feature selection and classifier applied on the various data sets. The most predictive model had an area under the receiver operating characteristic curve of 0.95, a sensitivity of 91%, and a specificity 92% in the training set; respective values in the validation set were 0.97, 91%, and 92%. In conclusion, MRI-based radiomics may be useful to stratify patients receiving neoadjuvant chemotherapy for osteosarcomas. Keywords: MRI, Skeletal-Axial, Oncology, Radiomics, Osteosarcoma, Pediatrics Supplemental material is available for this article. © RSNA, 2022.

Mapping Local Climate Zones in the Urban Environment: The Optimal Combination of Data Source and Classifier.

The novel concept of local climate zones (LCZs) provides a consistent classification framework for studies of the urban thermal environment. However, the development of urban climate science is severely hampered by the lack of high-resolution data to map LCZs. Using Gaofen-6 and Sentinel-1/2 as data sources, this study designed four schemes using convolutional neural network (CNN) and random forest (RF) classifiers, respectively, to demonstrate the potential of high-resolution images in LCZ mapping and evaluate the optimal combination of different data sources and classifiers. The results showed that the combination of GF-6 and CNN (S3) was considered the best LCZ classification scheme for urban areas, with OA and kappa coefficients of 85.9% and 0.842, respectively. The accuracy of urban building categories is above 80%, and the F1 score for each category is the highest, except for LCZ1 and LCZ5, where there is a small amount of confusion. The Sentinel-1/2-based RF classifier (S2) was second only to S3 and superior to the combination of GF-6 and random forest (S1), with OA and kappa coefficients of 64.4% and 0.612, respectively. The Sentinel-1/2 and CNN (S4) combination has the worst classification result, with an OA of only 39.9%. The LCZ classification map based on S3 shows that the urban building categories in Xi’an are mainly distributed within the second ring, while heavy industrial buildings have started to appear in the third ring. The urban periphery is mainly vegetated and bare land. In conclusion, CNN has the best application effect in the LCZ mapping task of high-resolution remote sensing images. In contrast, the random forest algorithm has better robustness in the band-abundant Sentinel data.

Situation Element Extraction Based on Fuzzy Rough Set and Combination Classifier.

Generalized network security situation awareness technology is divided into three processes: situation element extraction, situation understanding, and situation prediction. Situation element extraction is the most critical step in the whole process, and its extraction quality will directly affect the accuracy of situation understanding and prediction. In view of the shortcomings of current situation element extraction methods, this study makes an in-depth study on the network security situation element extraction algorithm and proposes a situation element extraction model based on the fuzzy rough set and combined classifier, which is used to improve the accuracy of situation elements acquisition, so as to provide a better data basis for situation understanding and prediction. In this study, the theory of fuzzy rough set is used to reduce the attributes of data without reducing the ability of data classification, which reduces the complexity of data; using the combination classifier theory and particle swarm optimization algorithm, a framework of situation element extraction is built, which can extract situation elements more accurately. The experimental results show that the network security situation element extraction framework proposed in this study can effectively shorten the extraction time of situation elements and improve the accuracy of situation element acquisition under the premise of ensuring the ability of data classification, thus proving the effectiveness and feasibility of the situation element extraction framework proposed in this study.

A DIGITAL PRINTING APPLICATION AS AN EXPRESSION IDENTIFICATION SYSTEM

Human Computer Interaction (HCI), a growing research field in science and engineering, aims to provide a natural way for humans to use computers as tools. Humans prefer to interact with each other mainly through speech, but also through facial expressions and gestures, for certain parts of the speech and displays of emotions. The identity, age, gender, and emotional state of a person can be obtained from his face. The impression we receive from the expression reflected on the face affects our interpretation of the spoken word and even our attitude towards the speaker himself. Although emotion recognition is an easy task for humans, it still proves to be a difficult task for computers to recognize user`s emotional state. Advances in this area promise to arm our technological environment by means for more effective interactions with humans, and hopefully the impact of facial expressions on cognition will increase rapidly in the future. Will do. In recent years, the adoption of digital has increased rapidly, and the quality has improved significantly. Digital printing has resulted in fast delivery and needs-based costs. This article describes a sophisticated combination classifier approach, an empirical study of ensembles, stacking, and voting. These three approaches were tested on Nave Bayes (NB), Kernel Naive Bayes (kNB), Neural Network (NN), Auto MultiLayer Perceptron (Auto MLP), and Decision Tree (DT), respectively. The main contribution of this paper is the improvement of the classification accuracy of facial expression recognition tasks. In both persondependent and nonpersondependent experiments we showed that using a combination of these classifier combinations gave significantly better results than using individual classifiers. It has been observed from experiments that the overall voting technique by voting achieves the best classification accuracy.