With increased use of Unified Payments Interface (UPI) transactions, fraudulent transactions have witnessed a sudden hike, which has been a welcome challenge for financial security. This project would develop an efficient fraud detection system with the help of machine learning and deep learning. Various classification techniques are proposed, ranging from Logistic Regression, K-Nearest Neighbors, Support Vector Machine, Naive Bayes, Decision Tree, Random Forest, XGBoost, and Convolutional Neural Network (CNN) to classify fraudulent transactions most efficiently. The data are subjected to intensive preprocessing, analysis, and then used to train the models with performance tested on the basis of accuracy, precision, recall, and F1-score. The best-performing model will be implemented in a web application with Flask, which is scalable, and fraud detection is possible in real-time. The primary aim of the research is to improve the security of electronic payment systems by offering an efficient and timely defense against financial fraud attacks.

Urban agriculture is faced with the big challenge of ensuring the best growing conditions and with minimal use of resources in a limited urban setting. The design and implementation of the IoT-Based Urban Agriculture Container Farm (IUACF) is a smart farming system based on the DHT-11, MQ-135, LDR, and ESP8266 microcontroller, which monitors air quality, light intensity, and temperature and humidity in real- time and provides alerts about the need to water plants and turn on lights, respectively. The information is sent to a cloud service, where it is supervised remotely with the help of automatic feedback in order to keep crops in the best possible state. The essence is to maximize the utilization of water and energy in urban container farming with the basic crops to include spinach, carrot, capsicum, and cucumber. The experimental outcomes indicate that IUACF realized 28 percent water and 15 percent increased crop production than traditional farming systems that used man power, and at the same time under 18 kWh/month of energy consumption is enough to operate a typical system. Modular design and wireless cloud integration enable the system to support smooth decision making that is data-driven and is scalable. These results show that the suggested IoT solution can significantly boost sustainability and efficiency in urban agriculture to offer a highquality framework on future Smart City food production programs.

The rapid proliferation of intelligent infrastructure has made the protection of homes ever more crucial. This Work describes a smart home security system based on AIoT, which combines sensing devices, cloud computing, and AI for real-time detection of invaders and subsequent alerts. The system uses a Raspberry Pi 3 connected with an ultrasonic sensor and a Pi Camera for activity tracking at access points. The camera takes a photograph once motion is detected, which is then sent to Amazon S3. An AWS Lambda function is then triggered for using Amazon Rekognition for comparison of identified faces with a collection of registered family members. Based on the success of this comparison, people are labeled as either known occupants or intruders. The results are stored in the cloud and immediately relayed to the homeowner through a Telegram bot sending marked up images and alerts. Experimental results suggest that the system attains on-average detection latency of 150 ms, a recognition accuracy of 93.7 % for faces, and alert dispatching within 1.2 seconds. Tests performed under different conditions of illumination, distance, and scenarios involving intruders demonstrate the robustness of the solution. With a serverless structure and modular components, this system offers a scalable, reliable, and cost-effective solution for smart home security.

Recruitment-related scams have increasingly posed serious concerns, often resulting in financial harm and emotional distress for job applicants. Conventional fraud detection techniques, which primarily rely on analyzing textual job content, tend to fall short when dealing with subtle language manipulations or adversarial changes. To address these limitations, this study introduces a custom-built Liquid Neural Network (LNN) model tailored for classifying job listings. LNNs are particularly suited for handling complex and variable language patterns due to their dynamic architecture. The detection pipeline includes essential preprocessing steps like tokenization, lemmatization, and the elimination of stopwords. It further explores feature representation using both TF-IDF (Term Frequency Inverse Document Frequency) and label encoding to assess their relative performance. The proposed LNN model achieved a balanced accuracy of 95.64%, outperforming conventional classifiers like Random Forest and AdaBoost. Experimental evaluation indicates that the proposed method reliably distinguishes fraudulent listings, offering an accurate and lightweight solution to mitigate recruitment fraud.

This study aimed to develop an emotion recognition model using brain signals. The brain computer interface (BCI) focuses on creating technology that enables direct brain-to-external device connections. There are two forms of BCI: invasive and non-invasive. In BCI, electroencephalography (EEG) is essential. EEG is a non-invasive method that involves applying electrodes to the scalp to capture electrical activity in the brain. EEG data is utilized to decode the user's intended emotions, activities, and thoughts. Emotions are important for human interaction, communication, and overall well-being. Many paralyzed people worldwide are unable to express their emotions or meet their needs, making it difficult to understand them, which leads to feelings of isolation. However, it is possible to detect emotions using BCI. Emotions are reflected in electrical brain activity and can be analyzed using EEG signals. The EEG signals are then decoded to detect a person's respective emotions. The decoding process mainly includes three steps. First, the signals are pre-processed to remove noise, and data is encoded. Second, the relevant features are extracted using the spectral power method. Third, emotions are classified using long short-term memory (LSTM), gated recurrent unit (GRU), and bidirectional long short-term memory (BiLSTM) algorithms. New EEG data is given to the model, and then emotions are displayed. The model developed using BiLSTM achieved an accuracy of 93.97%. A comparison was made with existing classification techniques that have used many three-dimensional (3D) models and the arousal-valence ratio to identify a person's emotion. The model's generalization will improve further by testing it on different types of datasets. The model's generalization improves further by testing it on different types of datasets.

Indoor positioning (IP) has attracted significant demand in diverse smart indoor consumer electronics applications like domotics appliances, automated energy management, patient tracking in hospitals, indoor navigation industries, etc. Most of these applications use Wi-Fi access points and a centralized server to create an IP network. The location of target nodes in these systems is disclosed, compromising user privacy. To overcome this, the current work proposes a federated KNN-based privacy-enforcing technique where the location coordinates of each target node are secured through discrete coordinate encryption. Hence the exact location coordinates are known only to the target node. However, real time consumer applications show that the privacy preserving technique increases localization error. But, efficient deployment of access points can improve localization accuracy. Therefore, we use Hausdorff distance to deploy dynamic access points based on the movements of the target nodes within convex hull regions. Experiments reveal that the proposed Hausdorff distance-based deployment model incorporated with a federated KNN results in better localization accuracy. Moreover, the proposed deployment technique does not require any additional hardware, making the system cost efficient.

The field of brain-computer interaction (BCI) is concerned with creating technology that enables direct brain-to-external device connection. Invasive and non-invasive BCI are the two different forms. In BCI, electroencephalography (EEG) is essential. EEG is a non-invasive method that involves applying electrodes to the scalp in order to capture electrical activity in the brain. To decipher the user's intended motions, activities, or thoughts, the EEG data is utilized. Emotions are important for human interaction, communication, and overall wellbeing. There are many paralyzed people throughout the world who are unable to express their emotions or meet their necessities. Hence, it is difficult to understand them and leads to feeling of isolation. But it is possible to detect the emotions using BCI. Emotions are reflected in the electrical brain activity and can be analyse using EEG signals. The EEG signals then decodes to detect the respective emotion of a person. Mainly three steps are included in the decoding process. First the signals are pre-processed to remove noise and data is encoded and second the relevant features are extracted using spectral power method, third classification of emotions using Bi directional Long Short-term Memory (BiLSTM).

A leading cause of chronic illnesses like diabetes and cardiovascular disorders, obesity is a serious global health concern. The goal of this research is to create a predictive framework for determining an individual's level of obesity based on their physical characteristics and eating habits. The suggested model provides excellent accuracy and resilience by utilizing ensemble learning approaches, particularly Random Forest and XGBoost classifiers. The dataset contains important variables that are analyzed to properly predict obesity levels, including BMI, frequency of physical activity, eating habits, and alcohol intake. Healthcare professionals can identify at-risk individuals and create focused intervention programs thanks to the system's insightful information. The results demonstrate the model's ability to produce accurate forecasts, paving the way for its use in useful healthcare applications to facilitate individualized intervention strategies.

With the increasing demand for machine learningin the medical field, this work addresses the growing need for accurate age prediction. Leveraging the NHANES dataset, which includes seven key features such as Body Mass Index and Glucose levels, the proposal utilizes advanced machine learning techniques to train a model that effectively predicts individuals' ages. After exploring various predictive approaches, a model was selected that demonstrated superior performance, highlighting its potential to enhance healthcare planning, risk assessment, and personalized health interventions. This data-driven approach promises to improve healthcare outcomes and support strategicdecision-making in the medical domain.

Indoor localization has attracted significant demand in diverse smart building applications like automated energy management, patient tracking in hospitals, industrial indoor navigation, etc. Most of the proposals use Wi-Fi access points to construct indoor localization systems and in such systems, the fundamental task is to deploy access points correctly. The existing literature has employed additional access points or related hardware to improve localization accuracy, which in turn results in expensive installation and maintenance costs. Our objective is to optimize deployment by modifying the positions of already existing access points without using any additional hardware. To achieve this, we propose a reverse multilateration based access point positioning framework that has three phases: the first phase uses multivariate regression to predict the coordinates of the target location based on received signal strength indicator values collected from multiple access points; the second phase identifies the misplaced access points using the cumulative error by distance ratio; and the third phase computes the new positions of access points through reverse multilateration. Experiments show that the proposal generates 888 correct predictions out of 960 data points, thereby improving the prediction accuracy by 4.79% when compared with existing methods.

Internet of Things (IoT) is steadily revolutionizing people’s lives, and accurate location sensing is crucial in achieving this. Global positioning system (GPS) is being widely used outdoors, but its accuracy decreases in indoor environments due to signal attenuation and multipath effect. Simultaneously, Wi-Fi fingerprint-based techniques that use signal strengths from Wi-Fi access points in a building have become more popular for performing indoor positioning. However, location-based services also result in smartphone’s battery life consumption because of frequent access point scanning. There are very few studies that focus on the energy conservation of localization systems, despite the fact that it is a significant factor in real-world applications. This paper proposes an intelligent scanning period dilation (ISPD) technique that uses a semi-centralized architecture and schedules Wi-Fi scans by allocating dynamic time intervals for each user. Experimental results show that the proposal saves 7.56% energy while reducing the location accuracy only by 1.35%.

IoT services are the basic building blocks of smart cities, and some of such crucial services are provided by smart buildings. Most of the services like smart meters, indoor navigation, lighting control, etc., which contribute to smart buildings, need the locations of people or objects within the building. This gave rise to Indoor Localization, where only the infrastructure of the building has to be used for localization as accessing the Global Positioning System is difficult in indoor environments. Many approaches have been proposed to predict locations based on the infrastructure available indoors, and some of such techniques use Wi-Fi access points. Still, unfortunately, very few studies have concentrated on tolerating faults while being cost-effective. This work discusses hardware implementation of indoor localization. It then proposes a learning algorithm SRNN (Speed Conscious Recurrent Neural Network) that uses the RSSI (Received Signal Strength Indicator) values of available Wi-Fi access points in the building and predicts the location. Also, fault-tolerant approaches termed nearest RSSI and the most recent RSSI using Kullback–Leibler Divergence have been proposed to improve the location accuracy when access points go down and are prone to faults. Both the proposed approaches nearest RSSI and most recent RSSI along with SRNN improve the location accuracy by 4% and 2.1%, respectively, over existing techniques and contribute to optimizing predicted location's accuracy in Indoor Localization an IoT service for smart buildings.

More buildings are becoming smart day by day which play a key role in development of smart cities and Internet of Things is essential in the development of such smart buildings. The incorporation of smart infrastructure is taken care right from the design phase of smart buildings itself. One of the crucial smart infrastructures in the development of smart buildings and smart cities is indoor localization. Being aware of the location or movement of people within a building can be very useful in several ways like energy management, location aware marketing services etc. and there are so many such IoT services. So the research problem here is to locate people within a building without using any additional infrastructure. Many research proposals have already been made over the last few years with the goal of predicting location in smart buildings but the real challenge lies with the accuracy of predicted location. User privacy and energy efficiency are also major challenges of indoor localization. Here we propose a random forest based machine learning algorithm that concentrates on improving the location accuracy in indoor localization as an IoT service for smart buildings. The obtained experimental results show 14% better success test prediction percentage in terms of overall deviation.

Indoor localization has become one of the most investigated techniques over the last decade due to the wide scale usage of smart phones and devices with wireless communication capabilities. This proliferates usage has made localization and user tracking much convenient for implementing IoT applications in smart buildings. Also due to the limitations of Global Navigation Satellite System in indoor environments, indoor positioning systems have gained much demand especially with various governments coming up with smart city and smart building initiatives. Various technologies like Wi-Fi, Bluetooth and Zigbee have been used to predict a user or device location indoor but the optimal location accuracy still remains a major problem. So, a review of some of the localization techniques being used and the challenges that could be faced while implementing a localization model is presented. Then the proposal comes up with a manual site survey approach which can be used to efficiently gather RSSI (Received Signal Strength Indicator) fingerprints from Wi-Fi Access Points so that the prediction models get trained well with the data collected and also presents some insights on the collected data.

An azimuth tree-based network algorithm is proposed on a uniform random distribution of scattered points which is actually graph-based. In the proposed azimuth on a tree-based network, first tree is established through the algorithm for finding the weight function of the tree would constrain the number of points to be considered during azimuth routing, thereby limiting the search space, reducing time complexities, and inducing further optimizations. The idea behind our paper would be applying the azimuth algorithm on a selected number of points, which would be achieved through the efficient tree-based routing protocol. So, firstly, we would apply our tree-based protocol on the uniform random distribution of points. The limited number of points output by this algorithm would be fed as input to azimuth routing. And in the protocol proposed by us, the aggregation of data in this tree-based network helps in reducing the network load and the energy consumption. And this research work mainly revolves around the criteria to be taken into consideration for balancing factors like these and construct a tree-based network which is better in terms of both lifetime of the network and the successful routing of protocols. With the help of simulation implemented using C++ and then constructing its corresponding graph, we have shown how considering all the factors possible has improved the performance of tree-based network. And finally, a comparative analysis is done where our proposed model is compared to the already existing traditional routing protocols namely AODV, DSDV, LEACH, Azimuth-based algorithm.