Theses supervised by Dr. Öğr. Üyesi Sefer Kurnaz
63 theses · Altınbaş University
Implementation of nested dilated-based trans-r2unet for segmentation and adaptive efficientnet with region attention mechanism to classify
Dental caries is a highly prevalent oral disorder, and the strategies of deep learning have been employed for diagnosing caries with large populations by employing RGB images. The conventional attention-aided image categorization approaches have the issues of feature underutilization and simple interference by irrelevant and background data. The timely recognition of dental caries is significant for performing treatments. For this objective, the bitewing radiography is employed to provide the initial detection of caries. The utilization of deep structured models with neural network approaches helps process the large number of images that have been experimented nowadays and provides promising functionalities. The computer-assisted smart vision approaches applied by image processing and machine learning mechanisms are required to eliminate these drawbacks. Hence, deep learning mechanisms have accomplished remarkable diagnosis efficacy in the radiology sector. Hence, a novel deep learning-based dental caries detection and classification framework is designed in this work. Initially, Cone Beam Computed Tomography (CBCT) images utilized for the detection and classification of dental caries are collected from the benchmark resources. Further, the collected images are provided for the image segmentation phase. Here, the developed Nested Dilated-based Transformer Recurrent Residual Unet (NDT-R2Unet) framework is used to attain the segmented images and these images are provided as the input to the dental caries disease classification phase. In this phase, an Adaptive Efficient Net with Region Attention (AdaENet-RA) is employed to classify dental caries among the individuals. Moreover, several parameters in developed AdaENet-RA are tuned using Arbitrary Updated Wild Geese Migration Optimization (AUWGO) and provide the dental caries disease classified outcomes. Later, various validations are executed in the developed framework to verify the effectualness of the suggested framework over the classical techniques.
Kablosuz sensör ağları için bulanık mantık tabanlı kümeleme optimizasyonu
Clustering is a powerful technique that organizes the process of the system to confirm network scalability, reduces the consumed energy, as well as realizes the network lifetime extension. The clustering routing algorithm is commonly utilized within wireless sensor networks (WSNs) due to its high-energy efficiency and scalability. The energy source in WSNs may be restricted to the battery's ability of the sensor nodes. As transmission energy is proportional to the distance that exists separating a sender to receiver, clustering in WSN can assist reduce energy usage. Data may be sent from any sensor node to corresponding Cluster Head (CH) for efficient information gathering. There is a need for maximizing the WSNs lifetime, optimization of network operation is crucial. The suggesting clustering algorithm is applied to enhance the lifetime of WSNs by selecting cluster heads (CHs) and forming clusters. Researchers have developed Fuzzy Logic (FL) to address the issue of overburdening Cluster Head (CH) during cluster formation in WSN. FL focuses on the CH efficiency, distributing load between sensor nodes for enhancing lifetime of network. The proposed protocol fuzzy logic based-clustering (FLB-CLS) outperforms FD-LEACH and OC-FCM, regarding network lifetime and throughput, also improving advance for networks that have higher node density. Each sensor node calculates the probability of CH using a distributed fuzzy inference system. For various network sizes and topologies, the results of simulation specify enhancements in energy efficiency as a result of network lifetime along with consumed energy balancing across sensor nodes. with concerning to first node dead along with total node dead, results demonstrate an average improvement.
Yapay zeka kullanılarak kullanıcı tüketim verilerine dayalı tüketici profillerinin sınıflandırılması
Over-The-Top (OTT) platforms have grown dramatically in popularity in recent years, giving consumers access to a variety of multimedia material. Understanding consumer behavior has therefore become essential for platform providers and advertising. In order to categorize users based on their consumption rate which may be classed as low, medium, or high this research suggests the construction of a user consumption categorization system based on machine learning (ML) as Decision Tree (DT) , Random Forest ( RF) , Regression Algorithm , k-Nearest Neighbour (KNN), Bayesian Algorithm (BA), Gradient Boosting (GB) and XGBoost Classifier . In order to tailor the data and get it ready for clustering, the research will employ data preprocessing techniques. Next, it will evaluate several ML classifiers to see which one is the most accurate at predicting the user's consumption rate. Each algorithm's limits will be investigated, and the system will be combined with data analytics and mining software. By offering a useful application for categorizing user consumption habits, this research will advance the fields ML and may be useful to OTT platform providers, advertisers, and data analytics experts.
Scen-scada security: an enhanced osprey optimization-based cyber attack detection model in supervisory control and data acquisition system using serial cascaded ensemble network
A significant role of Supervisory Control and Data Acquisition (SCADA) systems is supporting power system operation, where Information and Communication Technology (ICT) is adopted to interconnect the devices, and it increases system complexity. Because of the interconnection of SCADA systems, the complexity is increased, and there is a chance of cyber security vulnerabilities. In addition, the SCADA networks with legacy devices are affected by inbuilt cyber security deliberation that has provided severe cyber security vulnerable points. With the adoption of local-area networks and Internet Protocol (IP)-driven proprietary, malicious or unauthorized user accesses the information from outside sources, and hence, the SCADA systems are weakened by the elaborate attacks. SCADA systems need to deliberate the Denial of Service (DoS) and catastrophic failure as well as maloperation, which may subsequently compromise the stability and safety of the operations in the power system. Therefore, the pertinent priority in SCADA is to strengthen cyber security for guaranteeing reliable operation and also, the system stability is governed with respect to communications integrity. The smart grid features are used in the conventional Machine learning approaches for identifying cyber-attacks. Hence, implementing an efficient and accurate cyber-attack detection approach with less computational overhead is still a crucial research problem in SCADA. So, a novel and secure model for cyber-attack detection in the SCADA system using advanced deep learning techniques together with the heuristic algorithm is executed in this research work. The SCADA data are collected from various power grids. The features from these data are optimally selected and fused with the optimal weights in order to obtain the weighted optimal features. The weighted optimal feature selection is done with the aid of the Enhanced Osprey Optimization Algorithm (EOOA). These optimally selected weighted features are given to the Serial Cascaded Ensemble Network (SCEN) to obtain the final detection output. The developed SCEN is made with the cascading of Autoencoder, Dilated Bidirectional Long Short Term Memory (Bi-LSTM), and Bayesian classifier. The parameters in the SCEN are tuned using the executed IOOA. The final detection of the presence or absence of a cyber attack is evaluated by this SCEN. The performance and the effectiveness of the developed model are verified and contrasted by conducting various experiments.
Data mining and machine learning for cyber security intrusion detection
In spite of the quick development in information technology, securing computer and network resources still remains as a major challenge and concern for various organizations and researchers, particularly after the growth of networks and progress of technology. Implementation and designing intrusion detection systems are become very significant in network security. Intrusion detection is the fundamental tool of network security in struggling against malicious cyber attacks and unlawful network access. Since the continuously growing of attacks, it has been a technological challenge for an intrusion detection system (IDS) to successfully recognize known attacks and unknown attacks with insufficient training data. For that reason in the present study, an innovative contributions are implemented based on data mining and machine learning techniques for accurately and professionally detecting both known attacks and unknown attacks with inaccurate or insufficient training information. Faced with the increase of more advanced attacks targeting information systems, a defense system has become vital. An intrusion detection system provides a first line of defense. A intrusion detection system monitor events within an information system or in one of the organs of the information system The objective of this research project is to design a lightweight intrusion detection system using artificial intelligence techniques, in particular deep learning techniques. The neural network will be trained and tested with the NSL KDD dataset.
Control and management of solar PV grid using SCADA system
Solar radiation is plentiful in the Middle East, and it is one of the most sustainable renewable energy sources. On the other hand, our countries have yet to make it one of the most essential energy sources. We created a testing method based on SCADA simulation to achieve maximum voltage stability in the PV grid in this thesis. Two key advantages are the ability to be flexible and the speed with which testing are completed. Thanks to recent advancements in "smart sensors" based on smart grid technology, measurement instruments may now be changed to meet the size and kind of equipment being evaluated. The results indicate that the built open-source SCADA system functions optimally and precisely, and that it might be used as a variable control and monitoring system for PV grids with little modifications.
Elektronik muayenede hile tespiti için bir sistem tasarımı
An advanced and the good education system is the backbone of any country's progress. Only by having very valuable students in your country and gaining notoriety for their talents and efforts will you be able to establish a high international reputation. In order to achieve this, an education system must be fraudulent, so that unworthy students do not get the places they do not deserve. The objective of this study is to design a system to avoid fraud on the basis of eye movement in examination rooms. The system finds people from the scene and then identifies and recognizes them. The following step consists of eye detection and eye movement monitoring to examine the student or not in the fraud. The technique is widely used in educational and business establishments, wherever examinations have been carried out.
Ecological system: Stability and bifurcation
A three-dimensional prey-predator model is suggested to explain the interplay between prey and two age stages of the predator. The fear function is modeled in the prey population. It has also been hypothesized that the prey is strong enough to show anti-predator behavior and has the ability to kill the predator as a natural response to resistance to predation. The traditional Holling's disk Eq. (Holling type II) is modified to explain the prey consumption to involve an additional food (AF) for the predator, taking into consideration the handling time, and the search time for the prey. The existence, uniqueness, and boundedness of the Sol. of the proposed model are investigated. The local stability analyses are carried out after deciding all viable equilibrium points (EPs). The Lyapunov method is also used to investigate the global stability analyses for this model. The probability of local bifurcation (LB), along with Hopf bifurcation (HB), is studied. Moreover, to complete our study and confirm our analytical results, Numerical simulations are performed to investigate the global dynamics of the model and determine the set of control parameters.
Machine learning algorthims for URLs classification
Phishing is a technique used to collect sensitive data from a user (password or credit card information) for future misuse by posing as a trustworthy source. It often takes advantage of the user's gullibility in ways that the user will not detect at first look and, in the worst-case scenario, the attacker maintains the user's data without the user's awareness. Typically, the URL is the first and simplest piece of information we know about a website. As a result, it is logical to design algorithms for distinguishing harmful from benign URLs. Additionally, accessing and downloading the website's material may be time-consuming and involves the danger of downloading potentially hazardous information. Machine Learning techniques are used to train a model on a collection of URLs specified as a set of characteristics and then predict and categorize the URLs as benign or dangerous. This technology enables us to identify and avoid possibly dangerous URLs in the near future. We concentrated on the challenge of detecting malicious URLs using machine learning approaches in our thesis.
Blockchaın teknolojisiyle e-devlet hizmetlerinde organizasyonel birlikte çalışabilirliğin otomatik yapılması
The research is concentrated on the development of a smart system for the preservation of government records in smart E-Government benchmark repository. In literature, many existing preservation techniques and models have been discussed and presented with their detailed comparison as preservation rate of digital data growth has increased. Many western countries have already upgraded their paper-based systems to the smart system for preservation of government records including Turkey. In methodology, we used python language for the implementation of smart system application with E-Government benchmark to preserve archived paper-based records, it likewise delivers a roadmap for the innovative arithmetical environment which is trained using the automation based blockchain.. The blockchain model proposes construction that knows how to be utilized to relate household tasks and commitments in the interior of a confined structure. Major archives (i.e. open access, closed, restricted, proprietary) are creating certain assistance for safeguarding numerical objects and files in the E-Government benchmark dataset. Records accession, normalization, and transformation have also been performed on the records during the process of preservation to clean the records and format conversion. As a result, we achieved good preservation of most of the records from 2015 to 2025, the goal is to preserve most of the records it belongs to digital format. The distribution of records preservation for major archives in terms of paper-based and digitally preserved records from the year 2015 to 2025. The preservation of government records was recorded as very low in 2015 as much of the record archival was paper-based in all major repositories of the government. However, year by year the smart system tends to preserve the records in digital format from the old paper-based format, and by 2025 large number of records are being converted to the digital format with a whopping accuracy of 98.68% for all the records where the system was trained on 80% of data and tested on rest of the 20% of data. Having such a smart system can be very helpful for preserving the records of the government, and useful for understanding how records are being preserved and their functionality.
A QOS-aware self-updating intrusion detection system using reinforcement learning
Machine Learning (ML) techniques are being used for Intrusion Detection Systems (IDS) according to the rapidly developing attacking techniques that are being used by the intruders to compromise a computer network. Supervised approaches have shown the ability to evaluate and predict the type of the incoming packet faster than unsupervised methods. However, unsupervised methods have shown better ability to be updated when a new type of attacks is introduced to the network, as such an update requires only providing samples of the attack to the IDS, unlike the use of a supervised ML method, which requires additional training. Accordingly, a new IDS is proposed in this study, based on the use of Reinforcement Learning (RL) and has the ability to automatically recognize the patterns in the new types of attacks and block access of packets that match the recognized patterns. This recognition is based on the ability of RL to adapt to the changes in the environment by measures the rewards of the selected actions, which in the proposed method is measured based on the Quality of Service (QoS) being provided on the network. However, to avoid blocking all incoming traffic to maximize the QoS, hence, the reward value, the proposed method also includes the ratio of traffic allowed to access the network with the computations of the reward, so that, the best reward is achieved when all normal traffic is allowed access, i.e., allow all traffic that does not reduce the QoS. Several experiments are conducted to investigate the influence of the hyperparameters of the proposed method, as well as investigating the ability of the proposed method on adapting to changes in the attack traffic and protect the network. The results show that the proposed method has been able to achieve a high accuracy of 96.72% on traffic that contain attacks that have never been included in the training of the neural network of the RL agent. In contrast, a classification- based neural network has not been able to detect any of the packets that are part of these attacks, which illustrates the superiority of the proposed method.
Diagnose colon disease by feature selection based on artificial neural network and group teaching optimization algorithm
Multiple imaging modalities have substantially enhanced the diagnostic precision of contemporary medicine. Medical imaging has progressed to the point where an accurate diagnosis is achievable, and therapy can be commenced quickly. This article does an excellent job describing how to detect colorectal cancer. The recommended method for diagnosing plant diseases use a group teaching optimization algorithm to identify the most pertinent elements of an image. Multiple-layered trained neural networks are used to identify pictures as benign or malignant. The suggested technique for colon cancer diagnosis achieves a mean accuracy of 92.72%, sensitivity of 93.14%, and F1-Score of 94.26% when tested on the color data set. Kvasir picture data were classified with 96.42 percent accuracy, precision, sensitivity, plant disease, and F1-Score using the proposed method. Experiments demonstrate that the proposed strategy is superior to 3Layer for classifying photos of plant diseases. CNN There are currently numerous imaging modalities available for use in medical diagnostics. The four algorithms (DF, RF, CNN, and TFL). Keywords: plants disease, Group teaching optimization algorithm, Feature selection, artificial
Brain epileptic seizure diagnosis using electroencephalographic EEG signals
Monitoring frameworks employing image handling have gained significantly more attention among experts on the moving qualities like minimal costs, dependability, and flexibility toward merging with various innovations. Accordingly, the forest area fire area structure is arranged by creating revelation estimations in view of pictures and picture dealing with methods to segregate fire occasions in light of affirmation from Checking Focuses. In this examination task, the validation of a fire advance notice is supported by the proposed fire check assessment out. The fire check appraisal is proposed for the sales for fire picture and non-fire picture for the reduction of misdirecting cautions using picture managing moves close. A histogram evening out, RGB covering space, and YCbCr model-based structure is envisioned for data extraction from an information picture. The relationship of fire and non-fire symbolism is depicted using rule-based depiction. Histogram change gives the power makeover of fire covering pixels in the data picture, which is moreover overseen for the extraction of RGB parts. The erased RGB parts are sent off the change stage, where the image is exchanged over totally to a YCbCr covering model for secluding luminance from chrominance. Keywords: Machine Learning, EEG Signal, chronic disorder, Seizure Detection
Data mining and machine leaning methods for cyber security
There is drastic increase in needs of networking and data sharing in today's world. Such globalization of increased information technology and development there exists need of network security. Firewalls may provide some level of security but they never alert administrator for upcoming attacks. In order to find such abnormal behavior of network packets there is need of reliable detection system for improvement of efficiency and accuracy. Many research works focused on machine learning approach for enhancing the efficiency of the intrusion detection system and to detect malicious network activity automatically on the basis of network packet behaviors. The proposed model is designed using machine learning approach for detection of malicious activities of the network packets. For that CICDDoS2019 dataset is used. In this research, we proposed statistical methods are used obtain Z-scores, mean, median and mode for determining the significant features, training was performed for 80% of the dataset. The remaining 20% dataset was tested and validation using decision tree, K-NN and Gradient boosting algorithm. Keywords: Machine Learning, IDS, Cyber Attacks
Investigation of data mining for marketing purposes
To make sure that a product reaches the type of costumers that are more likely to buy it, in the last decades online Ad-based marketing started to bring more costumers for a certain product. However, in this work by using some information about a bank's costumers including their age and expected salary we were able to predict whether they will respond positively or negatively to our product's Ad by Firstly using Lineplot and Lmplot visualizations in python. And to prove that for more complexed datasets data visualization is not enough to predict the costumer's behavior, therefore, powerful machine learning models have been applied on the same dataset to make more accurate predictions, three deferent classification models have been used: NaïveBayes, KNN and SVM, these models showed accuracies of 88%, 90%, and 88.75%, respectively. Finally, predictions of new data that do not exist in the data set were predicted by the same models.
New automatic (IDS) in IoTs with artificial intelligence techniques
Intrusion detection in wireless sensor networks is a crucial task to ensure the security and reliability of IoT systems. Deep learning techniques have been demonstrated to be effective in various fields, including intrusion detection. Thus, this study proposes a novel framework for intrusion detection in wireless sensor networks using deep learning techniques.Three methods for developing a Network Intrusion Detection System (NID) for the Internet of Things (IoT) will be presented in this study. The first method involves training a deep learning model with optimization algorithms, such as the Biogeography-Based Optimization (BBO) algorithm, to enhance the security of IoT systems. The optimization algorithm helps in finding the best parameters for the deep learning model, leading to improved performance in detecting intrusions. The second method combines optimization algorithms with a classified as a feature selection tool to create a new NID. Feature selection is the process of choosing a subset of relevant features from the original dataset to improve the performance of the classifier. In this method, optimization algorithms are used to select the most relevant features, and then a classifier is trained on these features to detect intrusions. Finally, a Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) layers, along with the "Adam" optimizer, will be utilized to be training and make an evaluation data and handle any null values present in the dataset. CNNs are commonly used in image processing and have been demonstrated to be effective in various applications, including intrusion detection. LSTM is a type of Recurrent Neural Network (RNN) that is particularly well suited for handling sequential data. The "Adam" optimizer is a widely used optimization algorithm that helps to minimize the loss function in deep learning models. In conclusion, this study aims to propose a novel framework for intrusion detection in wireless sensor networks using deep learning techniques. The three methods presented in this study will provide a comprehensive solution for detecting intrusions in IoT systems, ensuring the security and reliability of these systems.
The obstacles of e-management in Iraq
The main differences discussed throughout this research were included in this thesis. Public sector agencies are expanding and improving their program offerings and delivery methods to meet rising demand. Technology may help these organizations decrease costs and improve delivery. The intended application of a technology solution may vary per company. May be used to automate internal research by using tiny repositories and e-mail programs, or to boost public infrastructure efficiency by implementing an integrated framework between firms and consumers. Businesses of all sizes, whether self-employed, medium, large, public, or private, utilize technology. This applies to public and private businesses. These enterprises represent education, health, commerce, marketing, tourism, and banking. This research investigates the link between common training and implementing E-Government. Patriarchal growth type organizations were linked with excellent performance, while traditional leadership institutions had less training owing to a less stringent educational norm. Because of this pressure, middle managers play a key role in adopting E-Government. Management positions are the principal target of e-government opposition for a number of reasons, including power loss and self-interest. This research also applies to government, indicating that some government initiatives, such as the Iraqi authority service (a program enforced by the government to replace foreign employees with local (Iraqi) people), has an influence on e-government. Such regulations made it harder for non-Iraqi citizens to earn a life. Employees in centralized social structures were less able to adopt E-Government than those in traditional management systems. Staff engagement in E-Government decision-making was greater in group learning institutions than hierarchical ones. Hierarchical companies' top-down, self-governing decision-making sped faster adoption, despite poor employee approval
A localization mechanism method in IoT using grasshopper optimization algorithm and DVHop algorithm
Nowadays, different types of computer networks such as Wireless sensor networks (WSNs), the Internet of things (IoT), and wireless body area networks (WBANs) transfer information, share resources, and process information. The IoT is a novel network which interconnects various smart devices and can consist of heterogeneous components such as WSNs for monitoring and collecting information. Characterized by specific advantages, the IoT contains different types of nodes, each with few sensors to collect environmental information on agriculture, ecosystem, search and rescue, conflagrations, etc. Despite extensive applications and high flexibility in the modern world, the IoT faces specific challenges, the most important of which include routing, energy consumption and localization. Localization leads to other network challenges and thus can be considered the most important challenge in the IoT. Localization refers to a process aiming at determining the positions and locations of objects lacking global positioning system (GPS) and needing to use the information of network sensors and topology to estimate their own positions and locations. The distance vector hop (DV-Hop) algorithm is a range-free localization technique, in which the major challenge is that the number of hops between two nodes is multiplied by a number that is the same for all nodes leading to a significant reduction in the localization accuracy. In this paper, a network node with no GPS determines the hops from three anchor nodes with GPS. The location of smart objectscan be then estimated according to distances from those anchor nodes. Thereafter, a few positions can be created nearby to mitigate the error. Then each position can be regarded as a member of the grasshopper optimization algorithm (GOA) to minimize the localization error. According to the results obtained from implementation of the proposed algorithm, it is characterized by a lower localization error than grasshopper optimization, butterfly optimization, firefly and swarm optimization algorithms. From the results we obtained that the proposed method has high performance than other method and the localization error for the FA, BOA, GOA (Proposed), PSO are DV-Hop are 1.940, 0.530, 0.416, 0.490, and 0.238 respectively.
A secure 5G data communication protocol based novel ai technique
With the development of the Internet and communication technology, the number of smart devices that can connect to the Internet is increasing day by day. These devices do not have the same security mechanisms that computers and servers have. Today, 5G networks are becoming increasingly popular for monitoring different types of critical environments. Due to the large number of 5G devices, the security of these networks and devices is an important concern. In this study, new method-based machine learning techniques presented to detect attacks in the 5G data communication environments. The proposed method combined unsupervised and supervised techniques to detect attacks. The data first analyzed using Boltzmann machines which high level features extracted. Then, the extracted features wired to the AdaBoost that classify the extracted features to the normal and abnormal classes.
An investigational FW-MPM-LSTM approach for face recognition using defective data
Facial recognition systems are listed as a biometric system, because they are directly related to the facial features and characteristics. They are also based on the principles of image processing, machine vision and sometimes machine learning. Face recognition systems may consider imperfect information from images. In this case, it is essential to provide a series of image reconstruction mechanisms for matching faces. In this paper a robust method was implemented and tested on the face recognition dataset based on the image's segmentation techniques. The proposed approach is that in the pre-processing phase, image should be enhanced. The image segmentation and reconstruction step is then followed by extracting the best facial features using features such as lips, eyes, cheeks and face area. This operation is based on fractal model and wavelet transform. Next, to train and test the system, the LSTM neural network is optimized using a method called Moore Penrose Matrix which named the MPM-LSTM. The results represent the proposed approach have better performance in comparison to recent methods. The performance accuracy rate for L-SVM, L-SVM-Wo, K SVM, K-SVM-Wo, CS, CS-Wo were obtained 98, 98.5, 95, 94.5, 98.1, and 98.3 respectively, while the proposed method is obtained as 99.58
Android malware prediction using machine learning
Increasing daily malware that exploits the Internet has become a severe threat. The manual malicious software (a.k.a., malware) identification is no longer valid and effective due to the high prevalence of malware. Thus, automatic behavior-based malware detection using machine learning techniques seems an effective solution. Various studies have proven the efficiency of machine learning to detect and classify malware files. In this research, several machine learning algorithms, including Decision Tree, Random Forest, Logistic Regression, SVM, and KNN, have been investigated to detect malicious software applications. For classification purpose, five classes, namely Adware, Benign, Ransomware, SMS Malware, and Scareware, have been used. Experimental results demonstrated that machine learning algorithms are practical and efficient for malware detection, and 71% accuracy can be achieved with the help of these algorithms.
0000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000
In this thesis, we analyse the efficacy of the 3LCDM technique in ultra-fast optical fibre networking settings. In comparison to a typical 51 Gb/s non return to zero (NRZ- OOK) system, the dispersion tolerance of a 51 Gb/s 3LCDM system is much larger, as shown by the results. There is an 81 ps/nm chromatic dispersion tolerance at a BER of 10- • 51 Gb/s in the upper tier, and a 98 ps/nm chromatic dispersion tolerance in the bottom tier. Compared to the normal NRZ values for 40 Gb/s, which are about 48 ps/nm, these are much higher values.
An early warning system for fires in hospitals and health centers via the internet of things to reduce human and material losses
A fire alarm circuit is an easy-to-build device that, upon detecting fire, sounds an audible alert (often a buzzer or siren). Smoke detectors and fire alarms are both components of security systems that play a significant role in early fire detection and the prevention of injuries and property loss. Fire alarms and smoke detectors must be installed in all public and commercial buildings. This includes schools, government buildings, and other institutions. There are many expensive and complex fire alarm circuits in the form of stand-alone devices, but we designed a very simple fire alarm circuit using common components like Arduino uno r3,Flame sensor ,Wifi module ,DHT11 temperature and humidity sensor and Pressure sensor gy-68,the LCD screen and the GSM module. It is powered by this project is designed to be a fire early warning system in hospitals and health centers via the Internet of Things to reduce human and material losses.
Prediction of weather using data mining techniques
Weather forecasting has become increasingly essential as a result of its uses in a variety of industries, including agriculture, energy companies, and everyday life. Weather prediction has become a real-time challenge for the world in the previous decade. Because of the ever-changing meteorological conditions, forecasting is becoming more difficult. Weather forecasting is the prediction of how the current state of the atmosphere will change in the future. Understanding the numerous contributing elements that produce weather variations is crucial for efficient weather analysis. The process of recording meteorological data such as wind direction, wind speed, humidity, rainfall, temperature, and so on is known as weather forecasting. Since machine learning techniques are more robust to perturbations, in this project we applied linear regression and LSTM to predict the weather such as temperature, rainfall etc. and compare both approaches and analyzed it. We used two different datasets for the same. Coming to result that we got from each approaches was quite amazing. In the linear regression approach, we got mean absolute error about 96.32 mm and 2.69 celsius when performing rainfall and temperature prediction respectively whereas in the deep learning approach, the mean absolute error was 0.002268 degree celsius, 0.003266 km/h, and 0.003069 Pascal when performing temperature, wind speed and pressure prediction respectively. We could clearly see the difference between the outcomes.