International Journal of Communication Networks and Information Security (IJCNIS)

AI-POWERED CROP YIELD PREDICTION & OPTIMIZATION SYSTEM

2026-03-21T07:30:54+00:00

The AI-Powered Crop Yield Prediction and Optimization System focuses on improving agricultural productivity by using machine learning techniques to assist farmers in making better crop selection and yield estimation decisions. Agriculture plays a crucial role in the global economy, and selecting the appropriate crop along with accurately estimating its yield significantly affects farmers’ income and productivity. Traditional farming decisions are generally based on experience, intuition, and seasonal knowledge, which may not always produce optimal results and can sometimes lead to poor crop selection and reduced yields. To ADDRESS this issue, the proposed system, AgriAI, utilizes data-driven decision-making through machine learning algorithms. The system analyzes important soil parameters such as nitrogen, phosphorus, potassium, and pH value along with environmental factors including temperature, humidity, and rainfall to determine the most suitable crop for cultivation and predict the expected yield. Supervised machine learning techniques are applied to process and analyze the agricultural data efficiently. Random Forest is used for crop recommendation, while XGBoost is implemented for yield prediction in order to achieve high accuracy and reliable results. To further enhance system performance, preprocessing methods and feature selection techniques are used to remove irrelevant data and optimize important agricultural attributes. The system is developed using Python-based machine learning libraries and deployed through a web-based platform that provides an interactive interface for users. FastAPI and Flask frameworks are used to create APIs for real-time prediction, while React is utilized for building a responsive user interface. Additionally, the system can be deployed on cloud platforms such as AWS, Google Cloud, or Firebase to ensure scalability and accessibility. Overall, the system aims to support farmers in selecting suitable crops, predicting production levels, improving agricultural efficiency, and promoting sustainable precision farming practices.

TRUST SCORE PREDICTION FOR ONLINE PLATFORMS USING INTERACTION PATTERNS

2026-03-21T07:33:39+00:00

Online platforms such as e-commerce websites and social networking systems increasingly struggle to maintain trust among users due to the rising presence of fake accounts, malicious bots, fraudulent transactions, and manipulated reviews. Traditional reputation and trust evaluation mechanisms mainly rely on explicit feedback such as star ratings, written reviews, and manual verification processes. However, these methods are often static, vulnerable to manipulation, and unable to accurately reflect a user’s real-time credibility in dynamic online environments. To address these limitations, this project proposes a machine learning–based Trust Score Prediction System that evaluates user credibility by analyzing implicit behavioral interaction patterns rather than relying solely on user-provided feedback. The system examines behavioral attributes including session duration, clickstream behavior, response latency, navigation paths, transaction frequency, and interaction consistency to determine the trustworthiness of users. By focusing on how users behave rather than what they claim, the system becomes more robust, adaptive, and resistant to fraudulent manipulation. The proposed framework follows a complete machine learning pipeline involving data preprocessing, feature engineering, feature selection, and model training. Random Forest is utilized to identify and select the most influential behavioral features, while XGBoost (Extreme Gradient Boosting) is employed as the primary prediction model due to its high predictive accuracy, capability to handle large and imbalanced datasets, and built-in regularization mechanisms that prevent overfitting. The performance of the proposed model is evaluated using standard classification metrics including Accuracy, Precision, Recall, and F1-Score to ensure reliability and effectiveness. Unlike conventional static trust models, this system dynamically updates trust scores based on recent behavioral activity, enabling early detection of suspicious or compromised accounts. Overall, the proposed approach demonstrates how machine learning–driven behavioral analytics can significantly enhance trust evaluation, strengthen platform security, reduce fraud, and improve the overall reliability of modern online ecosystems.

NDL-NET: NEONATAL RESPIRATORY DISTRESS DIAGNOSIS

2026-03-21T07:36:34+00:00

Neonatal Respiratory Distress (NRD) is a serious and potentially life-threatening condition that affects newborn infants and requires rapid detection and timely medical intervention to prevent severe complications or mortality. The project NDL-NET: Neonatal Respiratory Distress Diagnosis presents an intelligent deep learning-based diagnostic support system designed to assist healthcare professionals in the early identification of respiratory distress in neonates. The primary objective of this system is to develop an automated, reliable, and accurate solution that can analyze neonatal clinical data and identify abnormal respiratory patterns at an early stage. The proposed framework utilizes advanced neural network techniques to process medical datasets containing respiratory indicators and other relevant neonatal health parameters. A comprehensive data preprocessing pipeline is implemented to improve the quality of input data by performing operations such as data cleaning, normalization, and feature extraction, ensuring efficient and accurate model training. The deep learning model is trained using labeled datasets so that it can learn complex patterns associated with respiratory distress and provide predictive diagnostic outputs. The system follows a structured development methodology consisting of data collection, preprocessing, model development, training, validation, and testing phases. Various performance evaluation metrics, including accuracy, precision, recall, and F1-score, are used to measure the reliability and effectiveness of the model. The system is developed using modern machine learning frameworks and programming tools to ensure scalability, computational efficiency, and robustness. Additionally, the modular architecture of the system allows seamless integration with hospital information systems or medical diagnostic platforms. By enabling intelligent pattern recognition and automated clinical analysis, NDL-NET enhances early detection capabilities, reduces diagnostic delays, and provides an effective decision-support tool to improve neonatal healthcare.

STOCK MARKET TREND PREDICTION USING DATA SCIENCE

2026-03-21T07:38:35+00:00

The stock market is a complex and highly dynamic financial environment influenced by numerous factors such as historical price movements, economic indicators, geopolitical events, market sentiment, and investor behavior. Due to its volatile and nonlinear characteristics, predicting future stock market trends has always been a challenging task for investors, analysts, and researchers. Traditional statistical models and manual forecasting approaches often fail to capture the hidden patterns and temporal dependencies present in large-scale financial datasets. To address these challenges, this project proposes a Stock Market Trend Prediction System that utilizes advanced Data Science and Deep Learning techniques to forecast future stock prices with improved accuracy and reliability. The proposed system employs a Long Short-Term Memory (LSTM) network, which is a specialized architecture of Recurrent Neural Networks (RNNs) designed specifically for time-series data analysis. Unlike conventional regression models that treat each data point independently, LSTM networks are capable of remembering previous information over long sequences, allowing them to learn complex patterns and long-term dependencies in stock price data. The system uses historical stock market data including Open, High, Low, Close, and Volume values collected from financial data sources such as Yahoo Finance APIs. During the preprocessing phase, the dataset undergoes cleaning, normalization using MinMax scaling, and the creation of sequential time-window datasets, where past observations (for example, the previous 60 days of stock prices) are used to predict future trends. The LSTM model is then trained on this processed data to minimize prediction errors, which are evaluated using performance metrics such as Mean Squared Error (MSE) and Root Mean Squared Error (RMSE). Visualization techniques are also used to compare actual stock prices with predicted values to assess the accuracy of trend forecasting. Overall, the system demonstrates how modern deep learning models can effectively analyze financial time-series data and provide a scalable, automated, and efficient framework for stock market trend prediction and investment decision support.

WEATHER FORECASTING USING TIME SERIES MODELS

2026-03-21T07:41:22+00:00

Weather forecasting plays a crucial role in agriculture, disaster management, transportation, and environmental planning. Accurate prediction of weather conditions such as temperature, rainfall, humidity, and wind speed helps individuals, farmers, and organizations make informed decisions and reduce risks caused by unexpected climate changes. This project presents a weather forecasting system using time series models to analyze historical weather data and predict future atmospheric conditions. Since weather data is collected sequentially over time, time series analysis is an effective approach for identifying patterns such as trends, seasonality, and cyclic variations. The system utilizes statistical models such as Autoregressive (AR), Moving Average (MA), Auto Regressive Integrated Moving Average (ARIMA), and Seasonal ARIMA (SARIMA), along with deep learning techniques like Long Short-Term Memory (LSTM) networks to improve prediction accuracy. The project involves multiple stages including data collection from reliable meteorological sources, data preprocessing to handle missing values and noise, feature engineering to extract meaningful weather indicators, and model training to learn patterns from historical data. The trained models generate forecasts for future weather conditions, which are then evaluated using performance metrics such as Mean Absolute Error (MAE) and Root Mean Square Error (RMSE). The system also provides graphical visualization of predicted results to make the forecasts easy to understand for users. By integrating machine learning techniques with time series analysis, the proposed system improves forecasting accuracy and reliability compared to traditional methods. Overall, this project demonstrates how advanced predictive models can support better planning, improve agricultural productivity, and assist in environmental monitoring and disaster preparedness.

PREDICTIVE IDENTIFICATION OF AT-RISK LEARNERS IN SKILL-BASED LEARNING PLATFORMS.

2026-03-21T07:43:58+00:00

This project focuses on the predictive identification of at-risk learners in skill-based online learning platforms using machine learning techniques. With the rapid growth of digital education and Massive Open Online Courses (MOOCs), online platforms have made learning accessible and flexible for a wide range of learners across the world. However, one of the major challenges faced by these platforms is the high dropout rate, where many learners discontinue courses before completion due to lack of motivation, low engagement, time constraints, or insufficient academic support. To address this issue, the proposed system aims to develop an Early Dropout Risk Prediction System that can identify students who are likely to disengage from a course at an early stage. The system collects learner interaction data such as login frequency, course progress, video engagement, quiz attempts, assignment submissions, and forum participation. This data undergoes preprocessing steps including data cleaning, handling missing values, removing duplicate records, and transforming raw data into meaningful engagement features through feature engineering. These processed features are then used to train machine learning models such as Logistic Regression, Random Forest, and XGBoost to classify learners as “At-Risk” or “Safe.” The trained model predicts the probability of dropout using a test dataset and generates risk scores for individual learners. Based on these predictions, instructors can identify at-risk students and provide timely interventions such as personalized feedback, guidance, or additional learning resources. By enabling early detection and proactive support, the system improves learner engagement, increases course completion rates, and supports data-driven decision-making in modern skill-based online learning environments.

DELAYED DECISION PATTERN DETECTION IN DIGITAL PLATFORMS

2026-03-21T07:46:30+00:00

This project presents a real-time delayed decision pattern detection system for digital platforms using deep learning techniques, specifically the Long Short-Term Memory (LSTM) model. In modern web applications, understanding user behavior during an active session is essential for improving user experience, increasing conversion rates, and enhancing security. Traditional analytics tools mainly focus on retrospective metrics such as session duration, page views, or conversion outcomes, which fail to capture the temporal dynamics of user interactions. This project addresses that limitation by analyzing the time gaps between consecutive user actions, such as clicks, scrolls, and navigation events, to identify behavioral patterns. The proposed system treats these interaction delays as sequential time-series data and uses an LSTM-based neural network to learn temporal dependencies within user sessions. Based on the learned patterns, the system classifies user behavior into categories such as decisive users, hesitant users, and anomalous or bot-like users. The architecture integrates a React-based frontend for capturing user interaction events, a Node.js and Express backend for processing and session management, a MongoDB database for storing interaction sequences, and a Python-based machine learning service for real-time inference. By continuously analyzing interaction rhythms during an active session, the system can generate predictions with confidence scores and provide immediate feedback or adaptive responses. This enables platforms to offer proactive assistance to confused users, optimize interface design, and detect suspicious automated activity. The proposed approach is also privacy-friendly since it relies solely on timing data rather than personal or contextual information. Overall, this project demonstrates how temporal behavioral analytics combined with deep learning can transform conventional web analytics into an intelligent real-time system that improves both user engagement and platform security.

DECISION FATIGUE DETECTION USING APP USAGE PATTERNS

2026-03-21T07:49:06+00:00

Decision fatigue is a psychological condition that occurs when individuals experience mental exhaustion due to prolonged decision-making, leading to reduced concentration, slower responses, and poor judgment. Traditional techniques used to detect decision fatigue mainly depend on physiological sensors such as EEG devices or heart-rate monitors, as well as self-reported questionnaires. Although effective, these approaches are often intrusive, expensive, and unsuitable for continuous real-time monitoring in everyday environments. To overcome these limitations, this project proposes a non-intrusive and behavior-based system for detecting decision fatigue by analyzing smartphone application usage patterns. The proposed system collects and analyzesbehavioral indicators such as application switching frequency, scrolling behavior, session duration, typing delays, and response latency. These indicators serve as digital behavioral markers that indirectly reflect the cognitive and mental state of the user. A machine learning framework is implemented to process and classify these behavioral patterns. Feature selection is performed using the Random Forest algorithm to identify the most relevant attributes influencing fatigue detection. Subsequently, the XGBoost classification model is applied to accurately categorize users into three cognitive states: Alert, Moderately Fatigued, and Decision Fatigued. The system is designed as a lightweight and scalable software solution that can run on smartphones without requiring additional hardware sensors. Experimental evaluation demonstrates that analyzing app usage patterns can effectively identify behavioral changes associated with cognitive fatigue. By providing early detection and insights into mental exhaustion, the proposed approach can help individuals manage workload, improve productivity, and maintain better cognitive well-being. This research demonstrates the potential of behavioral analytics and machine learning in developing practical, cost-effective mental state monitoring systems.

PLANT LEAF IDENTIFICATION USING CNN WITH WAVELET-BASED FEATURE EXTRACTION AND PCA

2026-03-21T07:51:45+00:00

Plant leaf identification plays an important role in agriculture, biodiversity conservation, environmental monitoring, and botanical research. Traditional plant identification methods rely heavily on expert knowledge and manual observation of leaf characteristics such as shape, texture, color, and vein structure, which can be time-consuming and prone to human error. With the rapid development of artificial intelligence and image processing techniques, automated plant identification systems have become an effective solution to overcome these limitations. This study proposes an intelligent plant leaf identification system using Convolutional Neural Networks (CNN) integrated with Wavelet-Based Feature Extraction and Principal Component Analysis (PCA). Initially, the system performs image preprocessing operations including resizing, normalization, and noise reduction to improve image quality and ensure consistent input for further analysis. Wavelet-based feature extraction is then applied to capture important texture and frequency characteristics of leaf images, enabling the system to identify complex patterns such as vein structures and edges. To reduce computational complexity and remove redundant features, PCA is used as a dimensionality reduction technique that selects the most significant components while preserving essential information. The optimized feature set is subsequently fed into a CNN model that automatically learns hierarchical representations of leaf features and performs plant species classification. The proposed system is evaluated using performance metrics such as accuracy, precision, recall, and F1-score to assess its effectiveness. Experimental results demonstrate that the integration of CNN with wavelet-based features and PCA significantly improves classification accuracy and computational efficiency. The developed system provides a scalable, automated, and intelligent solution for plant species identification, which can support applications in smart agriculture, biodiversity monitoring, and digital botanical databases.

BEHAVIOUR BASED LOYALITY FRAGILITY PREDICTION

2026-03-21T07:54:05+00:00

Customer loyalty plays a critical role in the long-term success and profitability of modern businesses. However, traditional loyalty analysis methods mainly focus on historical purchasing patterns and often fail to identify early warning signs of customer disengagement. In many cases, customers who appear loyal based on past transactions may gradually lose interest in a brand due to factors such as competitive alternatives, reduced satisfaction, or changing preferences. This phenomenon, known as loyalty fragility, can result in sudden customer churn if it is not detected in time. The objective of this project is to develop a Behavior-Based Loyalty Fragility Prediction System that identifies customers who are likely to disengage in the future by analyzing their behavioral patterns. The proposed system uses machine learning techniques to analyze customer transaction data, engagement metrics, and interaction trends. Key behavioral indicators such as recency, frequency, and monetary value (RFM) are combined with temporal behavioral changes to measure loyalty stability. Customer segmentation is performed using clustering techniques to group customers based on their behavioral characteristics. A predictive model is then developed using advanced machine learning algorithms such as XGBoost to estimate the probability of loyalty decline and potential churn. The system provides loyalty scores and fragility scores for each customer, enabling businesses to identify high-value but at-risk customers at an early stage. By transforming raw customer interaction data into actionable insights, the proposed system supports proactive customer retention strategies. Ultimately, the project contributes to improved customer relationship management, increased customer lifetime value, and data-driven decision making in competitive business environments.