Hand Written Text Recognition Using Deep Learning

An analysis of machine learning models for sentiment analysis of Tamil code-mixed data

To determine if creating voting ensembles combining convolutional neural networks (CNN), support vector machine (SVM), and multi-layer neural networks (NN) alongside clinical parameters improves the accuracy of artificial intelligence (AI) as a non-invasive method for predicting aneuploidy. A cohort of 699day 5 PGT-A tested blastocysts was used to train, validate, and test a CNN to classify embryos as euploid/aneuploid. All embryos were analyzed using a modified FAST-SeqS next-generation sequencing method. Patient characteristics such as maternal age, AMH level, paternal sperm quality, and total number of normally fertilized (2PN) embryos were processed using SVM and NN. To improve model performance, we created voting ensembles using CNN, SVM, and NN to combine our imaging data with clinical parameter variations. Statistical significance was evaluated with a one-sample t-test with 2 degrees of freedom. When assessing blastocyst images alone, the CNN test accuracy was 61.2% (± 1.32% SEM, n = 3 models) in correctly classifying euploid/aneuploid embryos (n = 140 embryos). When the best CNN model was assessed as a voting ensemble, the test accuracy improved to 65.0% (AMH; p = 0.1), 66.4% (maternal age; p = 0.06), 65.7% (maternal age, AMH; p = 0.08), 66.4% (maternal age, AMH, number of 2PNs; p = 0.06), and 71.4% (maternal age, AMH, number of 2PNs, sperm quality; p = 0.02) (n = 140 embryos). By combining CNNs with patient characteristics, voting ensembles can be created to improve the accuracy of classifying embryos as euploid/aneuploid from CNN alone, allowing for AI to serve as a potential non-invasive method to aid in karyotype screening and selection of embryos.

Accurate and stable estimation of the state of health (SOH), which is one of the critical indicators to characterize the ability of lithium-ion (Li-ion) batteries to store and release energy, is critical in the stable driving of electric vehicles. In this paper, a novel SOH estimation method based on the aging factors of battery, which combines convolutional neural network (CNN), wavelet neural network (WNN), and wavelet long short-term memory (WLSTM) named CNN–WNN–WLSTM, is designed. The proposed CNN–WNN–WLSTM estimation scheme inherits both the fast convergence and robust stability of the WNN, as well as the ability of long short-term memory neural network (LSTM) to extract the time series features of the data; moreover, using CNN can make the proposed algorithm extract the data features from the original battery data automatically, and the WNN–WLSTM is then adopted to produce the final SOH estimation by exploiting the features from the CNN. To further speed and achieve global optimization, the RMSprop optimizer, instead of the usually used Adagrad optimizer, is chosen as the solver of the CNN–WNN–WLSTM network. Experimental results on data set from the NASA Ames Prognostics Center of Excellence show that the proposed algorithm can be commendably used for Li-ion battery health management by quantitative comparison with other commonly used machine learning methods, such as back-propagation neural network, WNN, LSTM, WLSTM, convolutional neural network–long short-term memory neural network (CNN–LSTM), and Gaussian process regression.

Deep learning models have recently been proven to be successful in various natural language processing tasks, including sentiment analysis. Conventionally, a deep learning model’s architecture includes a feature extraction layer followed by a fully connected layer used to train the model parameters and classification task. In this paper, we employ a deep learning model with modified architecture that combines Convolutional Neural Network (CNN) and Bidirectional Long Short-Term Memory (Bi-LSTM) for feature extraction, with Support Vector Machine (SVM) for Arabic sentiment classification. In particular, we use a linear SVM classifier that utilizes the embedded vectors obtained from CNN and Bi-LSTM for polarity classification of Arabic reviews. The proposed method was tested on three publicly available datasets. The results show that the method achieved superior performance than the two baseline algorithms of CNN and SVM in all datasets.

In this paper, we proposed the modified deep learning method that combined Convolutional Neural Network (CNN) and Kernel K-Means clustering for lung cancer diagnosis. The Anti-PD-1 Immunotherapy Lung dataset obtained from The Cancer Imaging Archive was used to evaluate our proposed method. From this dataset, we use 400 Magnetic Resonance Imaging (MRI) images that manually labeled consists of 150 healthy lung images and 250 lung cancer images. As the first step, all the data was examined through the CNN architecture. The flatten neuron of the feature map for every image resulted from the convolutional layers in CNN is gained and passed through the kernel k-means clustering algorithm. This algorithm then used to obtain the centroid of each cluster that determines the prediction class of every data point in the validation set. The performance of our proposed method was evaluated using several k values in k-fold cross-validation. According to our experiments, our proposed method achieved the highest performance measure with 98.85 percent accuracy, 98.32 percent sensitivity, 99.40 percent precision, 99.39 percent specificity, and 98.86 percent F1-Score when using RBF kernel function with sigma=0.05 in 9-fold cross-validation. Those performance improves 1.31% sensitivity, 1.12% accuracy, 1.11% F1-Score, 0.92% specificity, and 0.91% precision compared to when using 5-fold cross-validation. It is even obtained in less than 8 seconds for passing the dataset to the CNN model and 40 ± 0.77 seconds for examined in kernel k-means clustering. Therefore, it was proved that our proposed method has an efficient and promised performance for lung cancer diagnosis from MRI images.

Flood is a cataclysmic event that occurs far and wide by introducing extraordinary misfortunes. The floods additionally compromise the human life and the economy of the included nation likewise gets influenced. Neural Networks are generally connected in flood gauging and achieved great outcomes. The main objective of the project is to predict the flood situation and intimate about it to the people. The importance of predicting the flood is widely increased. The human life can get damaged due to floods; it can prevent loss to economy and human if it is predicted earlier. For prediction there are various methods but the most commonly used method is Neural Networks. This paper develops deep learning approach by combining Convolutional Neural Network (CNN) and Modified Particle Swarm Optimization (MPSO) for the prediction that the area is flooded or not. CNN extracts the attributes cataclysm in all respects productively. CNN is vigorous to shadow. CNN can get the attributes of calamity palatably and it additionally overpowers the misdetection or miscount by administrators, which further influence the handiness of fiasco help. MPSO is applied as optimizer to search for the optimal parameter values for the CNN training process.

Convolutional feature learning and Hybrid CNN-HMM for scene number recognition

This paper puts forward an approach that combines Convolutional Neural Networks (CNNs) and Genetic Algorithm (GA) to detect accurately and quickly the diseases that affect potato leaves swiftly and accurately. The model of the CNN is employed for automated feature extraction from the images of leaves which are pivotal in differentiating between the healthy and the infected leaves. Optimization of CNN hyper parameters, like learning rate, the number of layers, and dimensions of filters, is not only time-consuming but also challenging to find optimal values, and genetic algorithms for the optimal values of the parameters of the Convolutional Neural Network. A genetic algorithm uses an iterative search process that aims to optimize a population of models of Convolutional Neural Networks over several generations. It achieves this through the use of optimal models to perform crossover and mutation hence efficiently searching for the optimal configuration of hyperparameters. The above combine will improve the performance of the model which leads to better performance in detecting the variety of diseases attacking the potato leaves, such as early blight and late blight. In this paper CNN model utilizing a genetic algorithm attains an accuracy of 98.3% in 50 epochs.

Bolt loosening is a significant concern in structural health monitoring (SHM), and the nonlinear wave modulation (NWM) technique shows promise for its detection. However, temperature variations affect piezoelectric sensors, complicating detection. This study examines thermal stress effects on NWM-based damage detection by simulating a beam with boundary loosening under thermal stress. A temperature compensation method was developed to mitigate these effects. A multibolt structure was also analyzed using a hybrid deep learning model combining convolutional neural networks (CNNs) and Dezert-Smarandache Theory (DSmT) for multisensor data fusion. Results show that thermal stress significantly reduces classification accuracy. At 60 °C, the Damage Index dropped to zero, and CNN classification accuracy fell from 87.5% (at 25 °C) to 27.08%. Applying the compensation algorithm restored CNN accuracy to 87.27%. Statistical tests confirmed the compensated results at 60 °C were comparable to those at 25 °C. The DSmT-based fusion model achieved 98.92% accuracy on average, and even at 60 °C, maintained 98.80% accuracy with compensation for a multibolted plate. It combines the probabilistic outputs of multiple CNNs using the PCR6 rule, which redistributes conflicting evidence to improve decision reliability. This fusion strategy effectively handles uncertainty across sensors and significantly enhances classification robustness under temperature variations. In conclusion, thermal stress adversely impacts SHM damage detection, but the proposed compensation method effectively restores performance. Moreover, integrating multisensor fusion with DSmT greatly enhances accuracy, offering a robust solution for SHM in varying thermal environments.

Human Activity Recognition (HAR) based on data from wearable sensors has become an attractive research topic thanks to its applications in different fields such as healthcare and smart environments. Recently, the advancement of deep learning with capability to perform automatically high-level feature extraction has achieved promising results. However, the performance of the deep learning models depends deeply on the characteristics of the datasets such as the number of classes, the inter-similarity and intra-variation. Therefore, directly comparing these models has become difficult since a wide variety of experimental protocols, evaluation metrics, and datasets are employed. In this paper, for the first time, a systematic evaluation of several deep learning models for HAR from wearable sensors is provided. In particular, three models named Convolutional Neural Network (CNN) [1], DeepConvLSTM - a combination of CNN and Long Short Term Memory (LSTM) [2], and SensCapsNet - a Capsule Neural Network for wearable sensor-based HAR [3] were implemented and evaluated on three benchmark datasets that are 19NonSens, CMDFall, and UCI-HAR dataset. Moreover, to have an intuitive explanation of deep learning models, a visualization of features learnt from these models is given. The evaluation codebase and results will be made publicly available for community use.

The electric energy consumption prediction (EECP) is an essential and complex task in intelligent power management system. EECP plays a significant role in drawing up a national energy development policy. Therefore, this study proposes an Electric Energy Consumption Prediction model utilizing the combination of Convolutional Neural Network (CNN) and Bi-directional Long Short-Term Memory (Bi-LSTM) that is named EECP-CBL model to predict electric energy consumption. In this framework, two CNNs in the first module extract the important information from several variables in the individual household electric power consumption (IHEPC) dataset. Then, Bi-LSTM module with two Bi-LSTM layers uses the above information as well as the trends of time series in two directions including the forward and backward states to make predictions. The obtained values in the Bi-LSTM module will be passed to the last module that consists of two fully connected layers for finally predicting the electric energy consumption in the future. The experiments were conducted to compare the prediction performances of the proposed model and the state-of-the-art models for the IHEPC dataset with several variants. The experimental results indicate that EECP-CBL framework outperforms the state-of-the-art approaches in terms of several performance metrics for electric energy consumption prediction on several variations of IHEPC dataset in real-time, short-term, medium-term and long-term timespans.

The rapid growth of user density and heterogeneous service requirements in 5G and emerging 6G networks has intensified multi-user interference, particularly in power-domain non-orthogonal multiple access (PD-NOMA) systems. Conventional successive interference cancellation (SIC) techniques suffer from error propagation and performance degradation under imperfect channel conditions, limiting their reliability in practical deployments. This paper proposes a deep learning–based interference cancellation framework that combines Convolutional Neural Networks (CNNs) and Bidirectional Long Short-Term Memory (BiLSTM) networks for end-to-end signal detection in downlink NOMA systems. The proposed model exploits CNN layers to extract spatial interference features from complex-valued received signals, while BiLSTM layers capture bidirectional temporal dependencies caused by channel fading and symbol correlation. Unlike traditional SIC-based receivers, DL-NOMA-CancelNet directly reconstructs user signals without explicit channel estimation or manual interference subtraction. Simulation results obtained using MATLAB under Rayleigh fading and AWGN conditions demonstrate that our framework significantly outperforms conventional Least Squares (LS), Minimum Mean Square Error (MMSE), and single-network deep learning detectors. Specifically, this method achieves up to an 8–10× reduction in bit error rate (BER) across a wide signal-to-noise ratio (SNR) range, while maintaining low inference latency and moderate model complexity. These results confirm that our work provides a robust and scalable solution for interference mitigation in future 5G/6G downlink NOMA systems, particularly for ultra-reliable low-latency communications (URLLC) and massive machine-type communications (mMTC).

Postpartum depression (PPD) is a serious and under-diagnosed mental health condition affecting mothers after the birth of a child, with severe consequences for the well-being of the mother and the infant. Early identification is difficult because it is multifactorial, reported subjectively, and the interactions between biological, psychological, and social factors are complex. This research suggests an intelligent and reliable machine learning framework for early prediction of PPD using multidimensional maternal health data. The dataset contains the demographic, clinical, psychosocial, and psychological attributes obtained through structured bilingual questionnaires and Edinburgh Postnatal Depression Scale (EPDS) assessments for 64 postpartum women with 28 mixed-type features. Advanced preprocessing, feature selection, and class balancing using SMote were exploited to resolve the problem of data heterogeneity and imbalance. The proposed hybrid deep learning model combines Convolutional Neural Networks (CNN) for feature extraction at different levels, Bidirectional Long Short-Term Memory (Bi-LSTM) for modeling in time, and an attention mechanism to pay Attention to clinically relevant indicators. Model evaluation was done using a 10-fold CV based on accuracy, sensitivity, specificity, and ROC-AUC metrics. The results show the proposed model is superior to conventional approaches as it has 96.8% prediction accuracy, 96.1% sensitivity, 97.4% specificity, and ROC-AUC 0.98. Ablation analysis validates the role of each component (CNN, Bi-LSTM, Attention, feature fusion, and SMote), which has a significant contribution to the prediction performance. The study further includes explainable AI tools (SHAP and LIME) for improved interpretability for clinical applications. These results suggest that artificial intelligence-enhanced prediction of PPD can identify high-risk mothers early, facilitate early psychological interventions, and improve maternal mental health outcomes. The proposed framework provides a scalable and clinically relevant solution to supporting evidence-based maternal healthcare decision-making and could reduce undiagnosed PPD in a real-world setting.

In order to accurately predict the lane‐changing trajectory of the vehicle and improve the driving safety of the vehicle, a lane‐changing trajectory prediction model based on the combination of convolutional neural network (CNN) and bidirectional long short‐term memory (BiLSTM) neural network is proposed by comprehensively considering the historical driving behavior, the spatial characteristics of surrounding vehicles and the bidirectional time sequence information of the vehicle trajectory. Firstly, the vehicle trajectory data are filtered and smoothed, and it is divided into three categories: left lane change, right lane change, and straight driving, and a lane change trajectory sample set is constructed. Secondly, CNN‐BiLSTM model is constructed to identify the sample set of lane‐changing trajectory. Considering the interaction between vehicles in the driving process, the information of predicted vehicle, and surrounding vehicles is taken as the input of the model. The extracted feature vector is input to the BiLSTM layer for prediction after the CNN layer feature extraction, and the horizontal and vertical coordinates of the target vehicle at the next time are output. Thirdly, the trajectory data of the US‐101 dataset in NGSIM is selected to verify the performance of the CNN‐BiLSTM model, and at the same time, it is compared with models such as CNN‐LSTM, long short‐term memory (LSTM), BiLSTM, and CNN‐GRU‐Att. Finally, the verification result shows that the overall fitting degree of the vehicle lane change trajectory prediction of the proposed model reaches 99.50%, and the mean square error and mean absolute error are 0.0003076 and 0.01417, which are improved compared with other models. In the meanwhile, the research on multistep trajectory prediction in different prediction time domains is carried out. It was found that the longer the prediction time domain is, the lower the prediction performance of the model decreases, but the prediction accuracy still reached more than 96%, and it was able to accurately predict the lane change trajectory.

Forecasting The Air Temperature at a Weather Station Using Deep Neural Networks