Project Report
Browse
311 results
Search Results
Item Classification of Network Traffic Anomalies Using Deep Learning Techniques(2025-01-14) Khan, Md. Rabby HossainPredicting network traffic is essential to optimize network resource management, congestion avoidance, anomaly detection, and QoS in general. This paper presents and investigates a deep learning technique for the precise forecast of future network traffic through historical data. A dataset of 1,000,000 records and 11 features was used (700,000 samples for training and 300,000 for testing). The raw traffic data was divided into a normal class and an anomaly class for the classification and anomaly detection task. Data Preprocessing The steps of missing value treatment, normalization, data cleaning, and reshaping the data for meeting the minimum input requirement of the deep learning model were performed. Three deep learning algorithms (multilayer perceptron (MLP), feedforward neural network (FNN), and autoencoder (AE)) were developed and evaluated. They were chosen for their ability to model complex, non-linear relations from the network traffic and in order to obtain representations that traditional statistical models have not been able to learn. The experimental results showed that the MLP and FNN models produced high accuracy rates of 0.99, which was indicative of high predictive ability. The Autoencoder, despite its inferior performance with an accuracy of 0.94, also performed well in unsupervised learning and anomaly detection. Performance measures such as precision, recall, F1-score, confusion matrices, and ROC/Precision-Recall curves demonstrated the robustness and generalization of the models. The comparative study demonstrated that deep learning models, such as MLP and FNN, were more efficient than the conventional statistical predicting anomalies models. These results demonstrate the efficiency and scalability of deep learning for real-time network traffic prediction and anomaly detection, providing an intelligent and proactive network management technique for advanced communication systems. We deployed our top-performing model online and are currently examining the results produced visuallyItem Predicting Student Stress and Smartphone Addiction using Machine Learning(Daffodil International University, 2025-05-14) Akter, Sumyia; Akter, SadiaStudent stress and smartphone addiction have emerged as critical issues in contemporary academic environments, affecting mental health, academic performance, and overall well-being. This study explores the intricate relationships among behavioral factors, physiological indicators, and smartphone usage patterns using machine learning (ML) techniques. A total of ten regression models—Linear Regression, Decision Tree, Random Forest, Gradient Boosting, Support Vector Regressor (SVR), K-Nearest Neighbors (KNN), ElasticNet, XGBoost, LightGBM, and CatBoost— were evaluated for their ability to predict self-reported stress and addiction levels among students. Performance was measured using MSE, RMSE, R2, and computational efficiency. Results revealed that CatBoost demonstrated superior performance for stress prediction, achieving the lowest MSE (1.634) and highest R2 (0.793), while Linear Regression performed best for addiction prediction with the lowest MSE (0.377) and highest R2 (0.954). Correlation analysis highlighted strong associations between high stress levels and poor academic performance (r = 0.85), reduced sleep duration (r = –0.69), and high smartphone dependency. Notably, nighttime phone usage, frequent device unlocks, and high notification counts were found to significantly influence both stress and addiction levels. Beyond model accuracy, this study provides a comprehensive impact analysis across societal, environmental, ethical, and sustainability dimensions. It emphasizes the urgent need for proactive strategies in educational and mental health domains to mitigate digital overdependence and stress. Furthermore, it advocates for sustainable research practices, including energy-efficient computing and privacy-centered ethical frameworks, aligning technological progress with social responsibility. The findings pave the way for targeted interventions through mobile applications and policy initiatives to enhance student well-being in an increasingly digital academic landscape.Item Analyzing and Predicting the Social Media Addiction Using Machine Learning(Daffodil International University, 2025-01-12) Tuhin, Hemayet Hossain; Kayum, Md AbdulThe new reality of our modern life is social media. Modern science has given people a better civilization, increased the standard of living, but has taken away all the emotions of life. Social media has become an essential part of our modern lives. It significantly affects how people communicate, work, and interact with each other. Social media offers many benefits, such as increased connectivity and information sharing, but excessive use has led to growing concerns about addiction, which can negatively impact on mental health, productivity, and relationships. Our goal is to predict social media addiction and evaluate its impact on professional and personal life using machine learning algorithm. We used six machine learning classifiers like Random Forest, Extra Tree Classifier, Linear Discriminant Analysis (LDA), Gaussian Naive Bayes, and ensemble approaches including Stacking and AdaBoost to analyze a dataset of 522 samples that collected via structured questionnaires. Which included data from students and employed people. Outperforming the other models, Stacking Ensemble model achieved the highest accuracy of 91.71% in classifying social media addiction behaviors. This study not only provides valuable insights into identifying addiction behaviors, but it also proposes a practical solution in the form of a mobile application based on a superior predictive model. The goal of this application is to help people increase their consciousness and encourage people to adopt good social media practices.Item A study of Bangladeshi customers reviews sentiment analysis using BNLP and machine learning approaches(Daffodil International University, 2025-01-12) Sarker, Mst. SwarnaThis work concerns a sentiment analysis of the Bangladeshi customer reviews through bangla natural language processing and machine learning approaches. The main goal is to create a model that makes it possible to define whether customers have a positive or negative attitude toward a company. For training and testing the model 6,445 reviews of products or services have been collected and annotated for sentiment. Some of the preprocessing steps include cleaning the text data where noise such as stop words and any special characters were removed from the text data so as to allow for analysis. The analysis used text preprocessing methodologies such as TF-IDF to quantize the textual data into machine learning formats. Decision Tree, Random Forest, SVM, SGD, XGB Classifier used to identify the model with optimal performance in sentiment classification. The project also entailed designing the application programming interface using Streamlit and setting the function where users can enter the custom text and immediately get a sentiment analysis report. First outcomes indicate that the models attained high accuracy of sentiment prediction which proves the importance of using machine learning strategies in analyzing customer’s opinions. This paper emphasizes the role of NLP in analyzing consumer behavior and offers a useful instrument – the Customer Sentiment Analyzer – for businesses to evaluate customer’s opinions. It also contributes to the literature on processing regional language provides an understanding of the sentiment of Bangladeshi markets.Item Human security with supervised learning: Automatic hate speech detection with encoding method on violence features(Daffodil International University, 2025-01-12) Tanjin, SubrinaThis investigation goals to develop an intelligent system that identifies hate speech in audio recordings and replaces offending phrases with a beep sound while maintaining the speaker's natural voice quality. Feature extraction and noise reduction, especially Mel- frequency cepstral coefficients, are done through a dataset of over 3,000 voice samples of both hate and non-hate speeches, made possible by the Librosa package for effective audio processing. Various machine learning models, such as Random Forest, XGBoost, GBoost, KNN, and Logistic Regression, classify audio samples as hate or non-hate speech. It comes up to an incredible 85% detection accuracy. Wherever hate speech is detected, the deep learning capabilities ensure the system smoothly converts the objectionable words to a beep without influencing the overall tone and rhythm of speech. In days to come, real-time speech processing will also be developed whereby this system can mark and change speech during a live conversation. For the time being, the concentration remains on processing audio files. Furthermore, the integration of robust cybersecurity measures secures users' data in processing and storage with full compliance to privacy laws. Given its novelty in voice processing, this research incorporates a powerful method for moderating bad speech, with the opportunity to make digital communication platforms more inclusive, safe, and resistant to harmful material.Item Study of Social Media Sentiment Analysis Using Machine Learning Algorithm And NLP Approaches(Daffodil International University, 2025-01-13) Sarker, Md. NahidThe steady increase of the social networking sites resulted in the generation of large amounts of the user-generated text data that enable the use of sentiment analysis to gain insights into the general public sentiment. This project mainly concerns the construction of a sentiment classifier on text data from social media through the usage of machine learning and natural language processing. The process flow entails a heavy data pre-processing whereby the text is normalized, tokenized, de-stop worded, and lemmatized. For subjectivity and polarity scores, TextBlob is used to sort out informative comments based on their positive, negative or neutral sentiment. Finally, the feature extraction was done on the text data using Term Frequency-Inverse Document Frequency (TF-IDF) vectorizer to transform the text features into numbers. The multiple machine classifiers under consideration include Naive Bayes, Support Vector Machine (SVM) and Decision Tree. The highest accuracy model is compiled into an API using it for sentiment analysis on new entries made by users. The presented model shows high quality in terms of sentiment prediction therefore, the future work should concentrate on the integration of conventions NLP with machine learning algorithms. This project provides a solution of managing the huge volume of data collected from the SNS and analyzing the user sentiments for businesses and researchers. Potential future work is as follows One could use deep learning models to implement the process and where there is multilingual data the approach may have to be expanded further.Item Graph Neural Network Based Framework for Accurate Crop Disease Identification(Daffodil International University, 2025-01-12) Rahman, Md. MostafizurCrop diseases are a major challenge to agricultural productivity and hence diseases it at is the crucial earliest to identify to and avoid detect huge the loss of crops and ensure approach food for security. identifying the crop study diseases presents through a a new machine learning model in the (GNN). form the of work Graph is Neural divided Network into six phases to accomplish the paper’s objective, collecting the a first large of dataset which from entails a public domain that comprises images of crop diseases Powdery Mildew including and Ash Bitter Gourd Gourd Downy Mildew. These images are further divided into different disease categories to facilitate specific and examination. Further, employed several on image the processing techniques. In (GLCM) the with feature statistical extraction measures also with Co-occurrence deep Matrix learning extracted features using DenseNet121. Both kinds of features are critical for disease identification as as well they as provide high-level patterns. both detailed The most important contribution of this CropGNN study model is which the implements proposed graphs to represent the correlations between the features that have provides been an extracted. effective way for This Crop framework Disease Detection graph-based and learning. Through it enhancing integrates the state-of-the-art disease machine identification, learning this methods approach with enables better decision making in agriculture and helps in addressing one of the biggest challenges of our time which is sustainable farming and food security.Item Speech-Based Classification of Bengali Regional Accents using Machine Learning(Daffodil International University, 2025-01-13) Jahan, Naila Nushrat; Shomrat, Salman MahmudThis research investigates the classification of Bengali regional accents using speech data and machine learning techniques. Accurate recognition of regional accents plays a pivotal role in improving natural language processing systems in linguistically diverse regions such as Bangladesh. Speech data was collected from various regions, including 580 audio samples from Chandpur, 535 from General Bengali, 484 from Bogura, 456 from Chittagong, 420 from Sylhet, 413 from Barishal, and 28 from other areas. The dataset was preprocessed to extract key speech features, which were then used as inputs for machine learning models.Four machine learning algorithms were applied and evaluated: Random Forest, Decision Tree, K-Nearest Neighbors, and Logistic Regression. Among these, the Random Forest model demonstrated the highest accuracy, achieving 98.12%. The Decision Tree model followed with 87.67%, while K-Nearest Neighbors and Logistic Regression attained 75.17% and 65.92%, respectively. These findings highlight the superiority of ensemble methods such as Random Forest in managing complex and diverse datasets. The study also addresses the challenges in accent classification, particularly the variability in speech patterns and the limited data availability for less-represented regions. The inclusion of the "others" category further underlines the necessity of more comprehensive and balanced datasets to improve model generalizability. This work significantly contributes to the fields of computational linguistics and speech recognition, showcasing the effectiveness of machine learning in accent classification. The exceptional performance of the Random Forest model underscores its potential for real-world applications, such as automated transcription, accent-based recommendations, and language learning systems. Future work may focus on enhancing the dataset and leveraging advanced deep learning techniques to further improve accuracy and performance.Item Quality of ambient air prediction in Bangladesh - A Time Series Analysis and Machine Learning Approach(2025-01-12) Akhi, Habiba ChowdhuryAir quality is a key environmental concern, especially in rapidly urbanizing countries such as Bangladesh, where air pollution poses substantial dangers to public health and the economy. This study investigates air quality index (AQI) patterns and trends across 13 main cities in Bangladesh, strives to understand the elements that determine air quality, and attempts to predict future AQI levels using three predictive models: Linear Re- gression, ARIMA, and LSTM. The study focuses on the ability to estimate AQI values in diverse urban environments by analyzing these models based on many performance measures including mean squared error (MSE), root mean squared error (RMSE), mean absolute error (MAE) and R-squared (R2). This collection, containing historical AQI data for cities such as Dhaka, Chittagong, Rajshahi and Sylhet in Bangladesh, demonstrates considerable differences in air quality between locations. The results demonstrate that LSTM outperforms other models with consistent forecasts and reasonably low error met- rics across most cities. LSTM achieved the lowest MSEs and RMSEs in numerous urban regions, confirming its applicability for time-series forecasting of AQI. On the other hand, linear regression worked effectively in cities where AQI patterns were simpler and more linear, whereas LSTM, although a more advanced deep learning model, displayed some issues in managing non-linearity and seasonality in data. The relatively high MAE and low R2 of the model imply that more refining is required for its effective usage in AQI forecasting. The study also highlights seasonal and temporal tendencies, such as particular cities such as Dhaka and Narayanganj having higher pollution levels throughout certain months, underscoring the necessity for seasonal air quality management. This study shows the importance of predictive modelling in air quality monitoring and policy making. The results imply that specific efforts are needed for high-polluting cities, and future studies should consider hybrid models or more advanced machine learning approaches to boost the accuracy and usefulness of AQI predictions in Bangladesh all of this I display in my Website.Item Watermelon Leaf Disease Detection Using Machine Learning and Deep Learning Based Hybrid Approach(Daffodil International University, 2025-01-12) Rabbi, Razone ParvejWatermelon diseases significantly affect agricultural productivity, leading to economic losses and reduced crop quality. Traditional manual inspection methods are time-consuming, labor-intensive, and susceptible to human error. This research explores various Machine Learning (ML) and Deep Learning (DL) approaches to classify watermelon leaf diseases, with an emphasis on identifying the best-performing models for integration into a hybrid classification system. A dataset of approximately 5000 images, including healthy and diseased leaf samples, was sourced from Kaggle and preprocessed to ensure robust training. Initially, ML algorithms such as Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Decision Tree, and Random Forest were tested, yielding accuracies ranging from 59% to 88%. DL architectures, including baseline CNN, ResNet50, MobileNetV2, DenseNet121, and InceptionV3, were subsequently evaluated, with ResNet50 achieving the highest accuracy of 99.75%, closely followed by MobileNetV2 with 99%. Based on these findings, a hybrid model was constructed by combining SVM (for classification) and a pre-trained ResNet50 (for feature extraction), achieving an accuracy of 99.80%. This study demonstrates how artificial intelligence can be used practically to advance precision farming by promoting sustainable agricultural practices. By integrating ML and DL techniques into a hybrid model, this research contributes a significant step toward more accurate and impactful solutions for watermelon leaf disease classification, supporting sustainable agriculture and global food security.