In machine learning, self-supervised learning (SSL) has become a potent paradigm that allows models to acquire meaningful representations from unlabeled data without the need for laborious manual annotations. SSL bridges the gap between supervised and unsupervised learning by creating pretext tasks that produce supervisory signals straight from the data. A thorough analysis of self-supervised learning techniques is presented in this research, including important strategies including generative models, contrastive learning, and predictive learning frameworks. A critical analysis is conducted on prominent models such as SimCLR, MoCo, BYOL, and masked autoencoders. Additionally, a variety of applications in computer vision, natural language processing, speech recognition, and healthcare are examined in this study. Despite its success, SSL still has a number of issues, such as evaluation problems, computational cost, and representation collapse. Additionally, unresolved issues like robustness, scalability, and cross-domain generalization are covered. Lastly, prospective prospects for research are emphasized, such as multimodal learning and effective training techniques. An organized summary of the state of self-supervised learning now and its prospects for the future is given in this review.

In this study, improving the performance of Photovoltaic (PV) systems during low irradiance is addressed, with specific focus on the Chattogram Potenga area in Bangladesh. The efficiency of Photovoltaic (PV) systems depends on the level of irradiance and the efficiency of Maximum Power Point Tracking (MPPT) algorithms suffers much in low irradiance. In this paper, a novel approach is introduced to improve the performance of the P&O based MPPT technique, when the irradiance varies between 0 and $300 \mathrm{W} / \mathrm{m}^{2}$. The main contribution of the work is to provide a switching mechanism for multiple P&O's, where several P&O are dynamically switched according to different irradiance levels. We developed and optimized 16 P&O models designed for different irradiance ranges, which helps to enhance the system performance at various low-irradiance conditions. Results verify this strategy effectively improves the energy conversion efficiency, compared to the conventional MPPT methods. In addition, the presented method can increase the PV power generation in low irradiance and increase the reliability of the PV system in practical environment. This is of great importance for adaptive MPPT methods, which are commonly used to improve the performance of PV systems in cloudy regions.

Voice based applications are ruling over the era of automation because speech has a lot of factors that determine a speakers information as well as speech. Modern Automatic Speech Recognition (ASR) is a blessing in the field of Human-Computer Interaction (HCI) for efficient communication among humans and devices using Artificial Intelligence technology. Speech is one of the easiest mediums of communication because it has a lot of identical features for different speakers. Nowadays it is possible to determine speakers and their identity using their speech in terms of speaker recognition. In this paper, we presented a method that will provide a speakers geographical identity in a certain region using continuous Bengali speech. We consider eight different divisions of Bangladesh as the geographical region. We applied the Mel Frequency Cepstral Coefficient (MFCC) and Delta features on an Artificial Neural Network to classify speakers division. We performed some preprocessing tasks like noise reduction and 8-10 second segmentation of raw audio before feature extraction. We used our dataset of more than 45 hours of audio data from 633 individual male and female speakers. We recorded the highest accuracy of 85.44%.

The study was performed to assess the livelihood condition of fishing community nearby Dorsha River of Bangladesh. It was found that most of the fishermen belonged to the age groups of 31-40 years (30%), represented by 78.33% Married. Most of the fishing family (61.67%) is usually consisted of 4-5 members and they engaged in fishing for 6-10 (53.33%) years. Among the fishing community about 10% was illiterate, 23.33% was primary group and 61.67% was secondary group. About 46.07% of the fishermen received health advices from rural community doctors, 35% from upazila health complex and 18.33% got health service from MBBS doctors. The current survey showed that the average annual income by fishing was 13000±3000 BDT and mostly 51.67% are engaged in secondary jobs. The findings also showed that 91.67% of fishermen had electrical facilities and 86.67% drank safe water from tube well though they had homes made of wood and tin (70%). This assessment suggest that they need more institutional, organizational and technical and credit support for their socio-economic betterment and sustainable livelihood.

Potatoes are one of the world's most popular and economically important crops. For many uses in agriculture, breeding, and trading, accurate recognition of potato breeds is important. In recent years, deep learning algorithms have become effective tools for breed recognition tasks using pictures, which inspires researchers to explore their potential for recognizing potato breeds. The paper presents extensive research on the application of deep learning for potato breed recognition. The recognition of potatoes has been effectively performed using the five state-of-the-art deep learning models VGG16, ResNet50, Mobile-Net, Inception-v3, and a customized CNN. These models have been modeled to differentiate between several potato breeds based on their unique visual characteristics, such as size, shape, color, texture, and skin pattern, by being trained on images of various potato breeds. The performance of each of the deep learning models is evaluated through thorough evaluation and testing. Among the models, the customized CNN model gives the best accuracy. The customized CNN model's accuracy is 94.84%. We do not just evaluate the accuracy but rather some other indicative metrics, such as F1-score, recall, and precision, too.

AutoML (Automated Machine Learning) is an emerging field that aims to automate the process of building machine learning models. AutoML emerged to increase productivity and efficiency by automating as much as possible the inefficient work that occurs while repeating this process whenever machine learning is applied. In particular, research has been conducted for a long time on technologies that can effectively develop high-quality models by minimizing the intervention of model developers in the process from data preprocessing to algorithm selection and tuning. In this semantic review research, we summarize the data processing requirements for AutoML approaches and provide a detailed explanation. We place greater emphasis on neural architecture search (NAS) as it currently represents a highly popular sub-topic within the field of AutoML. NAS methods use machine learning algorithms to search through a large space of possible architectures and find the one that performs best on a given task. We provide a summary of the performance achieved by representative NAS algorithms on the CIFAR-10, CIFAR-100, ImageNet and well-known benchmark datasets. Additionally, we delve into several noteworthy research directions in NAS methods including one/two-stage NAS, one-shot NAS and joint hyperparameter with architecture optimization. We discussed how the search space size and complexity in NAS can vary depending on the specific problem being addressed. To conclude, we examine several open problems (SOTA problems) within current AutoML methods that assure further investigation in future research.

Bangladesh is facing a severe risk as a result of climate change, as there are six seasons in Bangladesh, each containing its circumstantial behavior. Agriculture is the main arm of this country for economic growth as ${2 8 \%}$ of people are farmers. The majority of citizens rely on rice for daily sustenance, which makes paddy a more important crop. But storms, rain, and drought are the main barriers to crop production in this country. Every year thousands of tons of paddy losses to adverse weather. In the era of Artificial Intelligence, we can make decisions by historical data. In this study, we will predict the yield of three major crops Aus, Aman, and Boro. For this purpose, we used 5 years (2019-2023) of weather and yield data for these crops. 11 different features are utilized in this work by 3 different deep learning architectures: ANN-based, LSTM-based, and LSTM + Ridge Regression-based hybrid model. The highest r2 score 0.90 with a mean squared error of ${0. 1 2}$ was achieved by our hybrid model 3.

About a hundred types of fresh vegetables are bought and sold in Bangladeshi local markets. The appearance of these vegetables seems to overlap in many cases, which leads urban people to get confused to recognize the vegetables properly. One instance of the most confusing vegetables is the gourd vegetables of the Cucurbitaceae family such as sponge gourd, ridge gourd, and snake gourd. Computer vision and deep learning can help in this regard through mobile or web applications. In this chapter, we have done exploratory work on gourd species recognition by applying deep learning to images of these three vegetables. The proposed approach begins with the collection of images of sponge gourd, ridge gourd, and snake gourd. A dataset of images of these three vegetables are prepared locally as well as from the Internet. After processing the images, we complete the training and testing phases using the convolutional neural network VGG-16. Then the hyperparameters corresponding to the VGG-16 model are tuned to come up with the best-configured model. An accountable accuracy of 99.5% has been achieved, which exhibits the prominence of VGG-16 in this problem domain.

Shrimp, the most popular shellfish in Bangladesh, is a good source of protein, minerals, vitamin D, and iodine that promote a healthy body and balanced nutrition. In Bangladesh shrimp is referred to as white gold. It consumes about 70% of exported agricultural food. In our country, about 56 species of shrimp are found. Most people do not know all of the species very well. Ordinary people even the fisherman are sometimes confused about different species because of looking like the same. To solve the problem in this work we introduced an intelligence mahine that can help people to concede Shrimp species accurately. We expect this work also help the export sector to differentiate the shrimp species monitoring. To achieve the goal, we build a custom CNN algorithm for image processing and feature extraction. We build three different CNN architectures and differentiate them by hyperparameter and number of convolutional layers. Model 1 and Model 3 both obtain an accuracy of 99.01%, however Model 3 was chosen as the final model for Computer Vision integration. Though both models generated the best accuracy why do we use model 3 as the final model? In this work, we will also describe with appropriate reason.

Using a customized approach, automatic classification of road surface condition and categorized data storing are proposed using DenseNet201. Road surface distress is one of the main issues affecting transportation safety. The first indication of a catastrophic asphalt pavement collapsing is a surface crack, that can later develop into a pothole and result in high repair costs. By replacing the surveillance system with the automated software program that we are recommending in this analysis, the traditional methods for identifying cracks or degradation in a road's surface, which involved manual examination by people, can be eliminated. DenseNet201 has outperformed other compared models with an accuracy of 98.75

Jujube make up a major portion of Bangladesh's total fruit production. It might be challenging to tell the differences between the many different species of jujube. The manual examination of jujube' physical qualities, which is time-consuming and prone to human mistakes, is the method of identification most commonly used in traditional methods. In this investigation, we make use of computer vision methods to zero in on particular jujube types that are native to the Bangladeshi region. In our approach, the question is solved with the assistance of a deep convolutional neural network (CNN) and Transfer Learning. Our method obtains an outstanding 98.0% accuracy on a test dataset after being trained on photographs of jujube taken in and around Bangladesh. Our work contributes to the growing body of research on applying computer vision and deep learning techniques to agricultural problems. Further research can be conducted to improve the accuracy of our system by collecting a larger dataset of jujube images, exploring the generalizability of our system to other regions and countries, and investigating the potential for using our system to recognize other fruit crops in Bangladesh or other countries.

The practice of automatically classifying images is gaining popularity. Many of us have very little knowledge of the local fruits, yet even in that case, we can vouch to their quality. In our study, we'll talk about a deep learning based system that can distinguish local fruits automatically. Fruit identification is a highly common activity, but automatically classifying fruits based on their placements, shapes, colors, and other attributes is a difficult task. Our study involved the collection of samples from several local areas, followed by the use of various transfer learning models, including VGG-19, Inception-v3, MobileNet, etc. MobileNet provided us with the highest accuracy of 99.53% among them. A top model was also suggested depending on the accuracy of our training. We used 60% of the image data from the 3240 total samples for training, for the purpose of validation we use 20% of the image data, and 20% of the total image data for testing. We received a satisfactory outcome after training and testing. Local fruits are classified as a consequence of this research model, which can be useful for everyday fruit identification.

Nowadays, gender Prediction has become a popular subject in machine learning and predicting gender by analyzing some text or names is very common while predicting gender by their taste or favourites is not so popular. As a result, for this paper, we established an aim of predicting people’s gender based on their preferences or requirements. There are a lot of choices and desires we want in our life partner, that’s why it was easy to detect gender on the basis of our choices. From ‘data storing’ to ‘selecting a model’ we have followed a modern workflow. We have made a public survey with proper questionnaires and encompass 758 data from different persons and tried to know their choices of choosing a life partner. We have tested our datum with 8 different Machine Learning Algorithms and from these algorithms, five algorithms-Gradient Boosting Classifier (GBC), Stochastic Gradient Classifier, eXtreme Gradient Boost (XGB), Decision Tree Classifier (DT), Random Forest (RF) comprises favorable accuracy from 90%-95.39%. The best correctness we have found is from the RF machine learning algorithm with 95.39% accuracy. The model can be a useful notion to apply in any Life Partner Chosen type applications (e.g., Wedding Service-Shaadi.com), according to the approach that arises from this study.

The modern education system is an essential part of the rise of technology. The E-learning education system is not just an experimental system; it is a vital learning system for the whole world over the last few months. In our research, we have developed our learning method in a more effective and modern way for students and teachers. For significant implementation, we are implementing convolutions neural networks and advanced data classifiers. The expression and mood analysis of a student during the onlineclass is the main focus of our study. For output measure, we divide the final output result as attentive, inattentive, understand, and neutral. Showing the output in real-time online class and for sensory analysis, we have used support vector machine (SVM) and OpenCV. The level of 5*4 neural network is created for this work. An advanced learning medium is proposed through our study. Teachers can monitor the live class and different feelings of a student during the class period through this system.

Recognition of medicinal plants is very important to enhance plant cultivation, boost production in the medical industry as well as protect the plant species from extinction. The plant leaf is a key feature in recognizing the plant. However, standard medicinal plant leaf data sets are scarce. This chapter deals with the development of a standard data set and also the recognition of plants from their leaves using a deep learning model, as deep learning models confirmed superior recognition accuracy. With this view, here, three benchmark deep convolutional neural networks (CNN), such as InceptionV3, MobileNet, and Xception are investigated to find their respective efficacy. Extensive experiments are performed using the developed data set to recognize 11 medicinal plants from their leaf images. MobileNet deep CNN architecture confirms the optimum performance based on four evaluation metrics derived from the confusion matrix.

There is a huge demand for the use of spices in South Asian countries. Each spice has a different taste, smell, and quality. People don't recognize spices and don't use them properly which results in not getting their actual qualities and wasting our time. Proper identification of spices is therefore required for use of proper spices. In this research, our proposed model can correctly recognize spices using computer vision and Convolutional Neural Networks (CNNs) with the help of images. there are 8377 images is used for training our custom build CNN architecture. For the best outcome, we have made three different models. And we choose the best model by applying different experimental theories. We obtained some remarkable outcomes from model 3 and, based on a test that which is never been seen before by our model and evaluation results selected the model that best performed. The highest accuracy reached 99.41% by model 3.

Bangladesh having a growing agricultural economy quick detection of plant leaf diseases is a primary requirement. Disease discovery with the existing diagnostic procedures demands a longer time. Therefore, growers frequently miss the best time for stopping and treating diseases. Further, early identification and classification of pumpkin leaf diseases extremely needed. This paper proposes to discover the pumpkin leaf diseases by utilizing a modern image processing procedure convolutional neural network (CNN). CNN applied for image classification and recognition because of its high accuracy. Besides, a comparison of traditional machine learning algorithms like support vector machines (SVM), K-nearest neighbor (KNN), decision tree, and Naive Bayes with the performance of CNN is demonstrated in our work. Tensorflow library was adopted to implement the CNN algorithm and Scikit-learn used in terms of utilizing the above-mentioned traditional machine learning algorithms. Finally, we detected the pumpkin leaf diseases by the algorithm that exhibits an assuring accuracy to our suggested approach.

Agoraphobia has become a fairly frequent disorder in today's environment. Agoraphobia is a spectrum of mental diseases characterized by strong dread and anxious feelings. The majority of individuals are completely oblivious of the problem. It's critical to detect it early on so that doctors can provide better treatment and avoid it from becoming a major issue. Recently, machine learning algorithms have been employed to analyze patient records in order to spot anomalies by modeling human thought or forming logical inferences. In this study, we strive to detect agoraphobia in the early stages. The basic theories and uses of machine learning algorithms in identifying anxiety kinds are also reviewed in this article. We primarily employed three feature selection techniques as well as a range of classification algorithms. The Naive Bayes, Random Forest, Decision Tree, KNN, and Support Vector Machine(SVM) are some of the categorization algorithms we utilize. After putting the Random Forest classification technique to the test, it was found to have greater accuracy of 98.02 percent than any other classification method that now is used.

Text classification is an essential and the most well-known topic of Artificial Intelligence as a discipline of Natural Language Processing (NLP). Because of the abundance of textual documents in Bangla, text classification has become a crucial subject. Natural Language Processing (NLP) in Bangla, at the same time, is not as developed as it is in English, and little study has been done in the context of the Bengali dialect, which is among the most widely used languages in the world. As a consequence, it's past time to address this issue in order to effective information management and data structure. The following is an example of a Bangla phrase from a narrative: assertive, interrogative, imperative, optative, or exclamatory text document. Numerous machine learning (ML) and deep learning (DL) algorithms are applied to categorize the sentence in the text document using the dataset. Our dataset is unique in that it was created by hand while keeping Bangla's sentence structure and origin in perspective. Within all the machine learning (ML) techniques, there are two that stands out: RN and DT provides the supreme exactness at 89.42%. As a deep learning strategy, between LSTM and RNN, LSTM exhibit superlative accuracy, with having an accuracy of 88.2 percent. Our experiment also offers a benefit in NLP for detecting the expression of textual data in the future execution, and hybrid approaches will be performed by increasing our dataset for improving the interaction between Bangla and the Natural language processing (NLP) field.

In the event that a student's success can be predicted, they may be able to improve their academic performance. The amount of instructional information available is increasing at a rapid rate. It is possible to use this information for educational purposes that will create an impact. Predicting student performance with Machine Learning is possible. It's possible that the new information may be put to good use in the classroom. Graduate students face a variety of challenges because of their low academic performance. Students' future performance may be predicted using historical data, and actionable suggestions for improvement can be provided. Daffodil International University's genuine mark data was likewise created using the program. After the data was collected, it was evaluated to see whether there were any data breaches. Remove unnecessary data from the dataset and the original needed data set comes out. The dataset may be analyzed using a variety of algorithms, including Decision Trees, Random Forests, Support Vector Machines, Gradient Boosters, Linear Regressions, and Neural Network Regressions. Most effective algorithm was selected to predict grade. After that, students were given tips and strategies for improving their grades.