Ensar Arif Sağbaş
Ensar Arif Sağbaş has a total of 9 co-authors. Here are the top co-authors by number of shared publications:
- Serkan Ballı
Department of Information Systems Engineering, Faculty of Technology, Muğla Sıtkı Koçman University, Muğla, Turkey
10 shared publications - Serdar Korukoglu
Ege Universitesi
4 shared publications - Musa Peker
Department of Information Systems Engineering, Mugla Sitki Kocman University, Muğla, Turkey
3 shared publications - Aybars Uğur
Ege Universitesi
1 shared publication Musa Peker
Muğla Sıtkı Koçman University
1 shared publication- Osman Özkaraca
Muğla Sıtkı Koçman University, Turkey
1 shared publication
Latest Publications
- 26 January 2027
- Published by IGI Global Scientific Publishing
- p. 107-154
Artificial Intelligence Applications for Precision Living
Artificial intelligence assisted image analysis is important for translating precision health and personalized well-being into dental practice. This chapter examines panoramic and periapical radiographs, cone-beam computed tomography (CBCT), intraoral and cephalometric images and digital surface models in terms of anatomical information, clinical applications and limitations. AI methods are reviewed from conventional machine learning and convolutional neural networks to object detection, segmentation, transformer-based architectures, self-supervised learning and foundation models. Applications including caries, endodontic and periodontal disease detection, orthodontic analysis, oral lesion screening, implant planning and tooth numbering are evaluated based on datasets, model architectures, validation strategies and performance metrics. The chapter emphasizes that high performance does not necessarily ensure clinical utility. External validation, calibration, prevention of data leakage, human oversight, privacy and equitable performance are essential for trustworthy AI in dentistry.
- 26 January 2027
- Published by IGI Global Scientific Publishing
- p. 107-154
Artificial Intelligence Applications for Precision Living
Artificial intelligence assisted image analysis is important for translating precision health and personalized well-being into dental practice. This chapter examines panoramic and periapical radiographs, cone-beam computed tomography (CBCT), intraoral and cephalometric images and digital surface models in terms of anatomical information, clinical applications and limitations. AI methods are reviewed from conventional machine learning and convolutional neural networks to object detection, segmentation, transformer-based architectures, self-supervised learning and foundation models. Applications including caries, endodontic and periodontal disease detection, orthodontic analysis, oral lesion screening, implant planning and tooth numbering are evaluated based on datasets, model architectures, validation strategies and performance metrics. The chapter emphasizes that high performance does not necessarily ensure clinical utility. External validation, calibration, prevention of data leakage, human oversight, privacy and equitable performance are essential for trustworthy AI in dentistry.
- Open Access
- 27 March 2026
- Published by Omer Halisdemir Universitesi in Ömer Halisdemir Üniversitesi Mühendislik Bilimleri Dergisi
- Vol. 16
Akıllı ulaşım sistemleri için çoklu veri türlerine dayalı hibrit sürücü davranışı sınıflandırma yaklaşımı
Akıllı ulaşım sistemlerinin gelişimiyle birlikte trafik güvenliği hâlâ kritik bir sorun olarak önemini korumaktadır. Bu çalışma, literatürde ilk kez OBD-II, akıllı telefon sensörleri, GPS ve çevresel ses verilerini bütünleştiren çoklu (multimodal) bir veri kümesi kullanarak sürücü davranışlarını sakin, normal ve agresif kategorilerine ayırmaktadır. Yeni oluşturulan bu çok kaynaklı veri kümesi, motor parametreleri, hareket sensörleri, yol özellikleri ve ortam seslerini kapsamaktadır. Farklı pencere uzunluklarında bölünen verilerden istatistiksel ve zaman–frekans tabanlı öznitelikler çıkarılmış, pencere boyutunun sınıflandırma doğruluğuna etkisi analiz edilmiştir. Support vector machine (SVM), Naïve Bayes (NB) ve multilayer perceptron (MLP) modelleri karşılaştırılmış; MLP en iyi bireysel sonuçları vermiştir. Bununla birlikte, çalışmanın özgün katkısı olan hibrit alt-uzay istifleme (hybrid subspace stacking) yöntemi, rastgele alt-uzay çeşitlendirmesi ve topluluk öğrenmeyi birleştirerek tüm modelleri geride bırakmış ve %95.69 doğrulukla en yüksek başarıyı sağlamıştır.
- 27 March 2026
- Published by Omer Halisdemir Universitesi in Ömer Halisdemir Üniversitesi Mühendislik Bilimleri Dergisi
- Vol. 16
Open AccessAkıllı ulaşım sistemleri için çoklu veri türlerine dayalı hibrit sürücü davranışı sınıflandırma yaklaşımı
Akıllı ulaşım sistemlerinin gelişimiyle birlikte trafik güvenliği hâlâ kritik bir sorun olarak önemini korumaktadır. Bu çalışma, literatürde ilk kez OBD-II, akıllı telefon sensörleri, GPS ve çevresel ses verilerini bütünleştiren çoklu (multimodal) bir veri kümesi kullanarak sürücü davranışlarını sakin, normal ve agresif kategorilerine ayırmaktadır. Yeni oluşturulan bu çok kaynaklı veri kümesi, motor parametreleri, hareket sensörleri, yol özellikleri ve ortam seslerini kapsamaktadır. Farklı pencere uzunluklarında bölünen verilerden istatistiksel ve zaman–frekans tabanlı öznitelikler çıkarılmış, pencere boyutunun sınıflandırma doğruluğuna etkisi analiz edilmiştir. Support vector machine (SVM), Naïve Bayes (NB) ve multilayer perceptron (MLP) modelleri karşılaştırılmış; MLP en iyi bireysel sonuçları vermiştir. Bununla birlikte, çalışmanın özgün katkısı olan hibrit alt-uzay istifleme (hybrid subspace stacking) yöntemi, rastgele alt-uzay çeşitlendirmesi ve topluluk öğrenmeyi birleştirerek tüm modelleri geride bırakmış ve %95.69 doğrulukla en yüksek başarıyı sağlamıştır.
- 3 March 2026
- Published by Burdur Mehmet Akif Ersoy University in Scientific Journal of Mehmet Akif Ersoy University
- Vol. 9 (1) , 94-108
A Low-Cost IoT Smart Pill Box with Physical Intake Verification: Design and Evaluation in Real-Life Use
Medication non-adherence undermines the treatment of chronic diseases, and reminder applications cannot show whether a dose was taken. This paper reports a 52-day field evaluation of a low-cost smart pill box that registers each intake event with an infrared sensor and a reed switch, time-stamps it with a battery-backed clock, queues events locally during network outages and forwards them to a Laravel server that alerts the patient, a caregiver and a physician by staged e-mails. One user with multiple sclerosis, who is also an author, used the device with three scheduled doses per day. Of 155 planned doses, 90 were taken (58.1%); excluding an 8-day period in which the device was deliberately switched off, adherence was 68.7% (95% CI 59–78%) and differed between phases (83.3%, 47.6% and 73.2%). Taken doses were registered at a median deviation of 0 min, 94.4% within five minutes of the alarm; e-mail reminders were followed by intake in 4 of 69 cases. All 90 events recorded after installation matched a planned dose, all 103 device packets were consistent with the database, the median device-to-server delay was 2 s, and the notification chain fired at its thresholds in every case. Deep sleep extended the runtime on a 10,000 mAh power bank by 2.6-fold (from 14 h to 36 h). The sensors register intake events but do not verify ingestion. The record was consistent, and the local alarm was a strong trigger for intake; consolidated caregiver notifications and peripheral power control are the next steps.
- 3 March 2026
- Published by Burdur Mehmet Akif Ersoy University in Scientific Journal of Mehmet Akif Ersoy University
- Vol. 9 (1) , 94-108
A Low-Cost IoT Smart Pill Box with Physical Intake Verification: Design and Evaluation in Real-Life Use
Medication non-adherence undermines the treatment of chronic diseases, and reminder applications cannot show whether a dose was taken. This paper reports a 52-day field evaluation of a low-cost smart pill box that registers each intake event with an infrared sensor and a reed switch, time-stamps it with a battery-backed clock, queues events locally during network outages and forwards them to a Laravel server that alerts the patient, a caregiver and a physician by staged e-mails. One user with multiple sclerosis, who is also an author, used the device with three scheduled doses per day. Of 155 planned doses, 90 were taken (58.1%); excluding an 8-day period in which the device was deliberately switched off, adherence was 68.7% (95% CI 59–78%) and differed between phases (83.3%, 47.6% and 73.2%). Taken doses were registered at a median deviation of 0 min, 94.4% within five minutes of the alarm; e-mail reminders were followed by intake in 4 of 69 cases. All 90 events recorded after installation matched a planned dose, all 103 device packets were consistent with the database, the median device-to-server delay was 2 s, and the notification chain fired at its thresholds in every case. Deep sleep extended the runtime on a 10,000 mAh power bank by 2.6-fold (from 14 h to 36 h). The sensors register intake events but do not verify ingestion. The record was consistent, and the local alarm was a strong trigger for intake; consolidated caregiver notifications and peripheral power control are the next steps.
- Open Access
- 13 January 2026
- Published by PeerJ in PeerJ Computer Science
- Vol. 12, e3493
Temporal and modal contributions to smartphone-based multimodal driving behavior classification: a comparative study of classical, deep learning, and patch-based time series transformer models
Understanding and classifying driving behavior is a critical component of modern intelligent transportation systems, with implications for traffic safety, fuel efficiency, and personalized driver support. As sensor-equipped mobile devices become increasingly pervasive, new opportunities have emerged for implementing data-driven behavior recognition systems in a cost-effective and accessible manner. This study presents a comprehensive and low-cost mobile framework for classifying driving behaviors using data collected entirely via a smartphone application. Unlike prior approaches that rely on embedded hardware, the proposed system performs all data acquisition and recording through a standard smartphone paired with a bluetooth-based on-board diagnostics II (OBD-II) adapter. The framework integrates multimodal sensor sources, including engine control unit (ECU) data, inertial motion sensors, enriched road metadata via the Overpass application programming interface (API), and environmental audio signals. Rather than isolating a single data domain, the system unifies mechanical, contextual, and behavioral dimensions to enable robust driving style analysis. Driving behavior was categorized into three classes (calm, normal, aggressive) using sliding time windows of 3, 5, 7, and 9 s. The effects of both window duration and data source composition on model performance were thoroughly evaluated. Classical machine learning models (artificial neural network (ANN), support vector machine (SVM), logistic regression (LR), Naive Bayes (NB)) based on engineered features were compared against deep learning architectures (convolutional neural network (CNN), long short-term memory (LSTM), gated recurrent unit (GRU), recurrent neural network (RNN)) trained on raw multivariate sequences. Results showed that multimodal integration substantially improved classification accuracy, with CNN achieving the highest performance. Additionally, the study incorporated patch-based time series transformer (PatchTST), a modern transformer-based architecture designed for time-series classification, across 128 experimental configurations. While CNN remained the top performer in overall accuracy, PatchTST yielded consistently stable and competitive results, particularly in long-window and feature-rich settings. Importantly, statistical analyses confirmed the significance of differences across feature sets, time windows, and model types. This included architectural parameters such as model depth and latent dimensionality, evaluated through analysis of variance (ANOVA) and post-hoc Tukey’s honestly significant difference (HSD) tests. By enabling high-accuracy driving behavior classification using only smartphone-based sensing, this study contributes a practical and scalable solution. The inclusion of attention-based PatchTST modeling further extends the methodological breadth, highlighting the role of transformer architectures in multivariate time-series analysis. Collectively, these contributions underscore the feasibility of deploying robust and intelligent driver monitoring systems in real-world environments.
- 13 January 2026
- Published by PeerJ in PeerJ Computer Science
- Vol. 12, e3493
Open AccessTemporal and modal contributions to smartphone-based multimodal driving behavior classification: a comparative study of classical, deep learning, and patch-based time series transformer models
Understanding and classifying driving behavior is a critical component of modern intelligent transportation systems, with implications for traffic safety, fuel efficiency, and personalized driver support. As sensor-equipped mobile devices become increasingly pervasive, new opportunities have emerged for implementing data-driven behavior recognition systems in a cost-effective and accessible manner. This study presents a comprehensive and low-cost mobile framework for classifying driving behaviors using data collected entirely via a smartphone application. Unlike prior approaches that rely on embedded hardware, the proposed system performs all data acquisition and recording through a standard smartphone paired with a bluetooth-based on-board diagnostics II (OBD-II) adapter. The framework integrates multimodal sensor sources, including engine control unit (ECU) data, inertial motion sensors, enriched road metadata via the Overpass application programming interface (API), and environmental audio signals. Rather than isolating a single data domain, the system unifies mechanical, contextual, and behavioral dimensions to enable robust driving style analysis. Driving behavior was categorized into three classes (calm, normal, aggressive) using sliding time windows of 3, 5, 7, and 9 s. The effects of both window duration and data source composition on model performance were thoroughly evaluated. Classical machine learning models (artificial neural network (ANN), support vector machine (SVM), logistic regression (LR), Naive Bayes (NB)) based on engineered features were compared against deep learning architectures (convolutional neural network (CNN), long short-term memory (LSTM), gated recurrent unit (GRU), recurrent neural network (RNN)) trained on raw multivariate sequences. Results showed that multimodal integration substantially improved classification accuracy, with CNN achieving the highest performance. Additionally, the study incorporated patch-based time series transformer (PatchTST), a modern transformer-based architecture designed for time-series classification, across 128 experimental configurations. While CNN remained the top performer in overall accuracy, PatchTST yielded consistently stable and competitive results, particularly in long-window and feature-rich settings. Importantly, statistical analyses confirmed the significance of differences across feature sets, time windows, and model types. This included architectural parameters such as model depth and latent dimensionality, evaluated through analysis of variance (ANOVA) and post-hoc Tukey’s honestly significant difference (HSD) tests. By enabling high-accuracy driving behavior classification using only smartphone-based sensing, this study contributes a practical and scalable solution. The inclusion of attention-based PatchTST modeling further extends the methodological breadth, highlighting the role of transformer architectures in multivariate time-series analysis. Collectively, these contributions underscore the feasibility of deploying robust and intelligent driver monitoring systems in real-world environments.
- Open Access
- 7 January 2024
- Published by Springer Nature in Neural Computing and Applications
- Vol. 36 (10) , 5433-5445
Machine learning-based novel continuous authentication system using soft keyboard typing behavior and motion sensor data
Smartphones utilize various authentication methods, including passwords, fingerprints, and face recognition. While this information is quite practical and easy to remember, it introduces several security issues. The primary concerns involve theft, password forgetfulness, or unauthorized password copying. Implementing behavioral biometrics for user authentication adds an extra layer of security. The main contribution of this study is the utilization of soft keyboard typing behavior, a behavioral biometric, for continuous user recognition. To achieve this, the phone's grip style and typing characteristics of users are scrutinized using data collected from motion sensors and the touchscreen panel. Another challenge in mobile device authentication pertains to recognition accuracy and processing time. To expedite and optimize data classification, a hybrid classification structure is suggested. This structure incorporates correlation-based feature selection and a straightforward logistic regression method, offering rapid and highly accurate classification outcomes—a further contribution of this study. Experimental results demonstrate that user identification can be accomplished in as little as 0.03 ms, with a classification accuracy of up to 93%. Continuous authentication systems offer greater security compared to one-time authentication systems. Nevertheless, these systems might not always yield the most precise results. Overcoming this challenge necessitates the development of an efficient software architecture. In line with this, an additional contribution of this study is an explanation of how to construct a continuous authentication system using the developed architecture.
- 7 January 2024
- Published by Springer Nature in Neural Computing and Applications
- Vol. 36 (10) , 5433-5445
Open AccessMachine learning-based novel continuous authentication system using soft keyboard typing behavior and motion sensor data
Smartphones utilize various authentication methods, including passwords, fingerprints, and face recognition. While this information is quite practical and easy to remember, it introduces several security issues. The primary concerns involve theft, password forgetfulness, or unauthorized password copying. Implementing behavioral biometrics for user authentication adds an extra layer of security. The main contribution of this study is the utilization of soft keyboard typing behavior, a behavioral biometric, for continuous user recognition. To achieve this, the phone's grip style and typing characteristics of users are scrutinized using data collected from motion sensors and the touchscreen panel. Another challenge in mobile device authentication pertains to recognition accuracy and processing time. To expedite and optimize data classification, a hybrid classification structure is suggested. This structure incorporates correlation-based feature selection and a straightforward logistic regression method, offering rapid and highly accurate classification outcomes—a further contribution of this study. Experimental results demonstrate that user identification can be accomplished in as little as 0.03 ms, with a classification accuracy of up to 93%. Continuous authentication systems offer greater security compared to one-time authentication systems. Nevertheless, these systems might not always yield the most precise results. Overcoming this challenge necessitates the development of an efficient software architecture. In line with this, an additional contribution of this study is an explanation of how to construct a continuous authentication system using the developed architecture.
- 30 September 2023
- Published by Academic Platform Journal of Engineering and Smart Systems in Academic Platform Journal of Engineering and Smart Systems
- Vol. 11 (3) , 95-106
Human Activity Recognition with Smartwatch Data by using Mahalanobis Distance-Based Outlier Detection and Ensemble Learning Methods
Recognition of human activities is part of smart healthcare applications. In this context, the detection of human actions with high accuracy has been a field that has been working for many years. With the increase in the usage of smart devices, smartphones and smartwatches have become the constant equipment of these studies thanks to their internal sensors. Sometimes abnormal data are included in data sets due to the way the data were collected and for reasons arising from the sensors. For this reason, it becomes important to detect outlier data. In this study, step counter and heart rate sensors were used in addition to an accelerometer and gyroscope in order to detect human activities. Afterward, the outliers were detected and cleared with a Mahalanobis distance-based approach. With the aim of achieving a better classification performance, machine learning methods were used by strengthening them with ensemble learning methods. The obtained results showed that step counter, heart rate sensors, and ensemble learning methods positively affect the success of the classification. In addition, it was found that the Mahalanobis distance-based outlier detection method increased the classification accuracy significantly.
- 30 September 2023
- Published by Academic Platform Journal of Engineering and Smart Systems in Academic Platform Journal of Engineering and Smart Systems
- Vol. 11 (3) , 95-106
Human Activity Recognition with Smartwatch Data by using Mahalanobis Distance-Based Outlier Detection and Ensemble Learning Methods
Recognition of human activities is part of smart healthcare applications. In this context, the detection of human actions with high accuracy has been a field that has been working for many years. With the increase in the usage of smart devices, smartphones and smartwatches have become the constant equipment of these studies thanks to their internal sensors. Sometimes abnormal data are included in data sets due to the way the data were collected and for reasons arising from the sensors. For this reason, it becomes important to detect outlier data. In this study, step counter and heart rate sensors were used in addition to an accelerometer and gyroscope in order to detect human activities. Afterward, the outliers were detected and cleared with a Mahalanobis distance-based approach. With the aim of achieving a better classification performance, machine learning methods were used by strengthening them with ensemble learning methods. The obtained results showed that step counter, heart rate sensors, and ensemble learning methods positively affect the success of the classification. In addition, it was found that the Mahalanobis distance-based outlier detection method increased the classification accuracy significantly.
- 28 March 2023
- Published by Firat Universitesi in Fırat Üniversitesi Mühendislik Bilimleri Dergisi
- Vol. 35 (1) , 239-250
Filtre Tabanlı Öznitelik Seçim Yöntemleri Kullanılarak Metinlerde Duygu Sınıflandırması Üzerine Karşılaştırmalı Bir Çalışma
Bir metin sınıflandırma problemi olarak duygu analizi, çevrimiçi metin belgelerinden öznel bilgi çıkarmanın kritik bir görevidir. Metin sınıflandırmanın önemli bir sorunu ise yüksek boyutluluktur. Boyut indirgeme, makine öğreniminde sınıflandırma performansını iyileştirmenin etkili bir yoludur. Alakasız özniteliklerin azaltılması eğitim süresini kısaltabilmekte ve sınıflandırma doğruluğunu artırabilmektedir. Farklı öznitelik seçim yöntemlerinin performansı, farklı veri kümelerinin özelliklerine bağlı olarak değişebilmektedir. Bu çalışmada filtre tabanlı 6 farklı öznitelik seçimi yönteminin (Korelasyon tabanlı öznitelik seçimi, Ki-kare, Kazanç oranı, Bilgi kazancı, OneR ve Simetrik belirsizlik katsayısı) performansı duygu sınıflandırmasında sıklıkla kullanılan 9 farklı veri kümesi üzerinde test edilmiş ve karşılaştırılmıştır. Bütün veri kümelerinde her bir öznitelik seçimi yöntemi için filtre puanları hesaplanmıştır. Elde edilen filtre puanları büyükten küçüğe sıralanmıştır. En yüksek filtre puanına sahip öznitelikten en düşük filtre puanına sahip özniteliğe doğru öznitelikler bir önceki alt kümeye eklenerek yeni alt kümeler oluşturulmuş ve sınıflandırılmıştır. Hesaplama sonuçları, önerilen yaklaşımın 9 genel duygu sınıflandırma veri kümesi için Çok terimli Naive Bayes sınıflandırıcısını kullanarak ortalama %94.34 doğruluk oranlarına ulaştığını göstermektedir. Arama uzayı dikkate alındığında, bu yaklaşımın geliştirilebilir ve mevcut yaklaşımlarla rekabet edebilir olduğu sonucuna varılabilir.
- 28 March 2023
- Published by Firat Universitesi in Fırat Üniversitesi Mühendislik Bilimleri Dergisi
- Vol. 35 (1) , 239-250
Filtre Tabanlı Öznitelik Seçim Yöntemleri Kullanılarak Metinlerde Duygu Sınıflandırması Üzerine Karşılaştırmalı Bir Çalışma
Bir metin sınıflandırma problemi olarak duygu analizi, çevrimiçi metin belgelerinden öznel bilgi çıkarmanın kritik bir görevidir. Metin sınıflandırmanın önemli bir sorunu ise yüksek boyutluluktur. Boyut indirgeme, makine öğreniminde sınıflandırma performansını iyileştirmenin etkili bir yoludur. Alakasız özniteliklerin azaltılması eğitim süresini kısaltabilmekte ve sınıflandırma doğruluğunu artırabilmektedir. Farklı öznitelik seçim yöntemlerinin performansı, farklı veri kümelerinin özelliklerine bağlı olarak değişebilmektedir. Bu çalışmada filtre tabanlı 6 farklı öznitelik seçimi yönteminin (Korelasyon tabanlı öznitelik seçimi, Ki-kare, Kazanç oranı, Bilgi kazancı, OneR ve Simetrik belirsizlik katsayısı) performansı duygu sınıflandırmasında sıklıkla kullanılan 9 farklı veri kümesi üzerinde test edilmiş ve karşılaştırılmıştır. Bütün veri kümelerinde her bir öznitelik seçimi yöntemi için filtre puanları hesaplanmıştır. Elde edilen filtre puanları büyükten küçüğe sıralanmıştır. En yüksek filtre puanına sahip öznitelikten en düşük filtre puanına sahip özniteliğe doğru öznitelikler bir önceki alt kümeye eklenerek yeni alt kümeler oluşturulmuş ve sınıflandırılmıştır. Hesaplama sonuçları, önerilen yaklaşımın 9 genel duygu sınıflandırma veri kümesi için Çok terimli Naive Bayes sınıflandırıcısını kullanarak ortalama %94.34 doğruluk oranlarına ulaştığını göstermektedir. Arama uzayı dikkate alındığında, bu yaklaşımın geliştirilebilir ve mevcut yaklaşımlarla rekabet edebilir olduğu sonucuna varılabilir.
- Open Access
- 1 January 2023
- Published by Afyon Kocatepe Universitesi Fen Ve Muhendislik Bilimleri Dergisi in Afyon Kocatepe University Journal of Sciences and Engineering
- Vol. 23 (5) , 1150-1159
Otokodlayıcı Tabanlı Boyut Azaltma ve Akıllı Saat Tabanlı Giyilebilir Hareket Algılayıcıları Kullanarak Yaşlılarda Düşme Tespiti
Falling is a serious health risk that can even result in death, especially for the elderly. For this reason, it is crucial to prevent falls and, in cases where prevention is not possible, to detect and intervene as soon as possible. Smartwatches are an ideal tool for fall detection due to their constant presence, rich sensor resources, and communication capabilities. The aim of this study is to detect falls in elderly people with high accuracy using motion sensor data obtained from smartwatches. To achieve this, a dataset was created consisting of falls and daily activities. Then, the feature vector was extracted which has provided successful results in signal processing studies. Afterward, the dimensionality of the dataset was reduced using an autoencoder-based approach in order to decrease the workload on smartwatches and ensure more accurate and faster classification. The dataset was classified using machine learning methods including naive Bayes, logistic regression, and C4.5 decision tree, and successful results were obtained. Their performances were then compared. It was observed that reducing the dimensionality had positive effects on both the classification accuracy and the computation time.
- 1 January 2023
- Published by Afyon Kocatepe Universitesi Fen Ve Muhendislik Bilimleri Dergisi in Afyon Kocatepe University Journal of Sciences and Engineering
- Vol. 23 (5) , 1150-1159
Open AccessOtokodlayıcı Tabanlı Boyut Azaltma ve Akıllı Saat Tabanlı Giyilebilir Hareket Algılayıcıları Kullanarak Yaşlılarda Düşme Tespiti
Falling is a serious health risk that can even result in death, especially for the elderly. For this reason, it is crucial to prevent falls and, in cases where prevention is not possible, to detect and intervene as soon as possible. Smartwatches are an ideal tool for fall detection due to their constant presence, rich sensor resources, and communication capabilities. The aim of this study is to detect falls in elderly people with high accuracy using motion sensor data obtained from smartwatches. To achieve this, a dataset was created consisting of falls and daily activities. Then, the feature vector was extracted which has provided successful results in signal processing studies. Afterward, the dimensionality of the dataset was reduced using an autoencoder-based approach in order to decrease the workload on smartwatches and ensure more accurate and faster classification. The dataset was classified using machine learning methods including naive Bayes, logistic regression, and C4.5 decision tree, and successful results were obtained. Their performances were then compared. It was observed that reducing the dimensionality had positive effects on both the classification accuracy and the computation time.