Fundamentals Of Neural Networks Architectures Algorithms And Applications

McCabe's cyclomatic complexity metric (1976) is widely cited as a useful predictor of various software attributes such as reliability and development effort This critique demonstrates that it is based upon poor theoretical foundations and an inadequate model of software development The argument that the metric provides the developer with a useful engineering approximation is not borne out by the empirical evidence Furthermore, it would appear that for a large class of software it is no more than a proxy for, and in many cases is outperformed by, lines of code<
 >

A critique of cyclomatic complexity as a software metric

Multi-sensor information fusion based on machine learning for real applications in human activity recognition: State-of-the-art and research challenges

Human Activity Recognition (HAR) employing inertial motion data has gained considerable momentum in recent years, both in research and industrial applications. From the abstract perspective, this has been driven by an acceleration in the building of intelligent and smart environments and systems that cover all aspects of human life including healthcare, sports, manufacturing, commerce, etc. Such environments and systems necessitate and subsume activity recognition, aimed at recognizing the actions, characteristics, and goals of one or more individuals from a temporal series of observations streamed from one or more sensors. Due to the reliance of conventional Machine Learning (ML) techniques on handcrafted features in the extraction process, current research suggests that deep-learning approaches are more applicable to automated feature extraction from raw sensor data. In this work, the generic HAR framework for smartphone sensor data is proposed, based on Long Short-Term Memory (LSTM) networks for time-series domains. Four baseline LSTM networks are comparatively studied to analyze the impact of using different kinds of smartphone sensor data. In addition, a hybrid LSTM network called 4-layer CNN-LSTM is proposed to improve recognition performance. The HAR method is evaluated on a public smartphone-based dataset of UCI-HAR through various combinations of sample generation processes (OW and NOW) and validation protocols (10-fold and LOSO cross validation). Moreover, Bayesian optimization techniques are used in this study since they are advantageous for tuning the hyperparameters of each LSTM network. The experimental results indicate that the proposed 4-layer CNN-LSTM network performs well in activity recognition, enhancing the average accuracy by up to 2.24% compared to prior state-of-the-art approaches.

https://www.mdpi.com/1424-8220/21/5/1636/pdf

LSTM Networks Using Smartphone Data for Sensor-Based Human Activity Recognition in Smart Homes

In the last decades, time series classification (TSC) has emerged as one of the most challenging issues in data mining, and extensive studies have been done on various methods, including algorithm-based and learning-based techniques. Sensor-based human activity recognition (HAR) is a TSC issue that has become one of the most sought-after fields among business and academia specialists because of the proliferation of smartphone technology and wearable movement sensors. Conventional approaches to feature extraction provide a significant challenge in feature selection. Deep learning is an efficient strategy in the HAR scientific field and has solved the issue of feature selection. Nevertheless, several obstacles remain to study topics, including classifier interpretation. This article integrates well-known deep learning methods, namely convolutional neural networks and RNN-based models. The new approach proved to be more effective than the existing state-of-the-art approach. We assessed our network on the multivariant time-series benchmark (UCI-HAR) and revealed that our model surpasses other models in terms of training time and overall accuracy.

Time Series Classification Using Deep Learning for HAR Based on Smart Wearable Sensors

Activity-based wellbeing administration integrating health apps on intelligent wearables is becoming increasingly prominent. The accessibility of multimodal sensors in everyday wearable devices makes it feasible to measure human behavior and provide wellness services that are aware of the user's environment. In smartphones, smartwatches, and many other wearables, integrated sensors provide a wealth of information that can be utilized to recognize human movement. This research shows how effective deep learning algorithms can extract abstract information for human movement categorization in daily life. We came up with an approach to deal with this issue is what we name one-dimensional deep residual neural network (1D-ResNet). With an overall accuracy of 94.15%, we observed that the proposed 1D-ResNet outperforms standard deep learning models in categorizing complicated human activities such as reading, writing, eating, and other hand-oriented tasks.

Recognition of Complex Human Activities for Wellness Management from Smartwatch using Deep Residual Neural Network

Currently, during holidays and weekends in urban areas, a huge number of shoppers can be seen crowding into shopping malls and supermarkets. When there are special offers and huge discounts, the number of people increases even more. In supermarkets these days, consumers need a shopping cart or basket when making their purchases of various items. In addition, the process of procuring the products is rather complicated as the shoppers must carry a basket every time in order to find the products they want and place them into the basket, followed by needing to deal with the calculation of the expenses. After completing their shopping list, the consumers must wait in a long line so that they can check out and pay the bill. As a result, in order to solve this problem, this research presents the development of a smart basket that is used for shopping. All of the products in the supermarket normally have a barcode tag. Therefore, a barcode reader of android smart phone will be included on the smart basket. In this way, the name and the cost of each item will be shown on the display of the mobile phone when the customer scans and places the products in the basket. The total bill will be added up to give the total cost of all of the purchases and stored to the memory of the microcontroller. The product information regarding the items contained in the basket will be transferred to the main computer via a transmitter. Moreover, validation of the accuracy of the shopping process will be performed by a weight sensor system built into the basket. Thus, waiting in line to check out and pay the bill while having to estimate the total cost will be avoided.

Design and Implementation of the Smart Shopping Basket Based on IoT Technology

The advancement of inertial sensor technology and the growing prevalence of intelligent wearables (including smart-watches, smart bands, and other intelligent devices) have aided in applying science and technology to automated Fall Detection Systems (FDSs). Over the past decade, wearable-based FDSs has garnered considerable scientific attention. In this context, ma-chine learning (ML) techniques have demonstrated remarkable efficacy in differentiating between falls and typical motions or Activities of Daily Living (ADLs) using the data recorded by wearable inertial sensors. Unfortunately, in most research, the effectiveness of machine learning classifiers was restricted by feature extraction and selection processes that relied on human-crafted decisions. A deep neural network was established to en-hance the capacity of fall detection, which combines convolutional layers and aggregated residual transformation in this study. The TensorFlow framework trained the proposed model to identify and categorize 15 special events: three distinct falls and twelve ADLs. On a publicly available benchmark dataset (UMAFall), the proposed model was analyzed and compared to existing baseline models. The findings indicate that the proposed model is superior to other models, with an accuracy of 96.72%.

Automatic Fall Detection using Deep Neural Networks with Aggregated Residual Transformation

Human activity recognition (HAR) plays an increasingly vital role in several industrial applications, including medical services and rehabilitation surveillance. With the fast growth of information and communications technology, wearable technologies have recently triggered a new human-computer interaction. Wearable inertial sensors (IMUs) are commonly used in the area of HAR because this data source provides the most insightful motion signal data. Lately, HAR studies have examined the enhancement of activity recognition using bio-signals like Electrocardiogram (ECG) and Photoplethysmography (PPG). Nevertheless, current HAR research was constrained by machine learning techniques that relied on human-crafted feature extraction. This research proposed a deep learning technique to effectively identify physical activity behaviors using ECG, PPG, and IMU sensor data. ResNet-SE is a deep residual network that incorporates convolutional processes, shortcut connections, and squeeze-and-excitement. We trained and evaluated baseline deep learning models to assess the suggested network, including the proposed model, using the public HAR dataset called Wrist_PPG dataset. According to experimental findings, the suggested method earned the most fantastic accuracy of F1-score. In addition, our results indicate that the PPG data can be utilized to classify physical workouts.

Sakorn Mekruksavanich

Papers

Time Series Classification Using Deep Learning for HAR Based on Smart Wearable Sensors

Recognition of Complex Human Activities for Wellness Management from Smartwatch using Deep Residual Neural Network

Design and Implementation of the Smart Shopping Basket Based on IoT Technology

Automatic Fall Detection using Deep Neural Networks with Aggregated Residual Transformation

Classification of Physical Exercise Activity from ECG, PPG and IMU Sensors using Deep Residual Network