CityLightSense: A Participatory Sensing-based System for Monitoring and Mapping of Illumination levels

Adequate nighttime lighting of city streets is necessary for safe vehicle and pedestrian movement, deterrent of crime, improvement of the citizens’ perceptions of safety, and so on. However, monitoring and mapping of illumination levels in city streets during the nighttime is a tedious activity that is usually based on manual inspection reports. The advancement in smartphone technology comes up with a better way to monitor city illumination using a rich set of smartphone-equipped inexpensive but powerful sensors (e.g., light sensor, GPS, etc). In this context, the main objective of this work is to use the power of smartphone sensors and IoT-cloud-based framework to collect, store, and analyze nighttime illumination data from citizens to generate high granular city illumination map. The development of high granular illumination map is an effective way of visualizing and assessing the illumination of city streets during nighttime. In this article, an illumination mapping algorithm called Street Illumination Mapping is proposed that works on participatory sensing-based illumination data collected using smartphones as IoT devices to generate city illumination map. The proposed method is evaluated on a real-world illumination dataset collected by participants in two different urban areas of city Kolkata. The results are also compared with the baseline mapping techniques, namely, Spatial k-Nearest Neighbors, Inverse Distance Weighting, Random Forest Regressor, Support Vector Regressor, and Artificial Neural Network.

A Genetic Algorithm and RNN-LSTM model for Remaining Battery Capacity Prediction

Li-ion batteries have diversified applications in everyday life. The temperature change, overcharging, over-discharging is playing critical roles in affecting battery life in a significant manner. In this paper, the deep learning-based method is applied for the prognostics of a single Li-ion battery. The proposed design uses a recurrent neural network variant, Long short term memory. The model's parameters are optimized through a Genetic Algorithm based parameter selector The method applies to a sequence of data values comprising of the voltage, the charge capacity, the current, and the temperature. The estimation of battery capacity is not only based on the current or defined state of the battery; instead, it is generated on the complete data profile. The robustness of the model is tested by comparing with techniques such as Support vector regressor, Kalman Filter, neural networks on normal and noisy test sets. The paper also proposes a feature selection and engineering scheme for battery capacity prediction. The proposed model outperforms the techniques available in literature with high generalization to noise and other perturbations. The model is independent of the section of charging curve used for prediction of battery capacity. Various experimentation has been conducted on the model and the results have been validated.

Machine-Learning-Enabled Virtual Screening for Inhibitors of Lysine-Specific Histone Demethylase 1

A machine learning approach has been applied to virtual screening for lysine specific demethylase 1 (LSD1) inhibitors. LSD1 is an important anti-cancer target. Machine learning models to predict activity were constructed using Morgan molecular fingerprints. The dataset, consisting of 931 molecules with LSD1 inhibition activity, was obtained from the ChEMBL database. An evaluation of several candidate algorithms on the main dataset revealed that the support vector regressor gave the best model, with a coefficient of determination (R2) of 0.703. Virtual screening, using this model, identified five predicted potent inhibitors from the ZINC database comprising more than 300,000 molecules. The virtual screening recovered a known inhibitor, RN1, as well as four compounds where activity against LSD1 had not previously been suggested. Thus, we performed a machine-learning-enabled virtual screening of LSD1 inhibitors using only the structural information of the molecules.

A geostationary lightning pseudo-observation generator utilizing low frequency ground-based lightning observations

Coincident Geostationary Lightning Mapper (GLM) and National Lightning Detection Network (NLDN) observations are used to build a generator of realistic lightning optical signal in the perspective to simulate Lightning Imager (LI) signal from European NLDN-like observations. Characteristics of GLM and NLDN flashes are used to train different machine learning (ML) models, that predict simulated pseudo-GLM flash extent, flash duration, and event number per flash (targets) from several NLDN flash characteristics. Comparing statistics of observed GLM targets and simulated pseudo-GLM targets, the most suitable ML-based target generators are identified. The simulated targets are then further processed to obtain pseudo-GLM events and flashes. In the perspective of lightning data assimilation, Flash Extent Density (FED) is derived from both observed and simulated GLM data. The best generators simulate accumulated hourly FED sums with a bias of 2% to the observation, while cumulated absolute differences remain of about 22 %. A visual comparison reveals that hourly simulated FED features local maxima at the similar geolocations as the FED derived from GLM observations. However, the simulated FED often exceeds the observed FED in regions of convective cores and high flash rates. The accumulated hourly area with FED>0 flashes per 5 km×5 km pixel simulated by some pseudo-GLM generators differs by only 7% to 8% from the observed values. The recommended generator uses a linear Support Vector Regressor (linSVR) to create pseudo-GLM FED. It provides the best balance between target simulation, hourly FED sum, and hourly electrified area.

Self-Perceived Loneliness and Depression During the COVID-19 Pandemic: a Two-Wave Replication Study

COVID-19 studies to date have documented some of the initial health consequences of lockdown restrictions adopted by many countries. Combining a data-driven machine learning paradigm and a statistical analysis approach, our previous paper documented a U-shape pattern in levels of self-perceived loneliness in both the UK and Greek populations during the first lockdown (17 April to 17 July 2020). The current paper aimed to test the robustness of these results. Specifically, we tested a) for the dependence of the chosen model by adopting a new one - namely, support vector regressor (SVR). Furthermore, b) whether the patterns of self-perceived loneliness found in data from the first UK national lockdown could be generalizable to the second wave of the UK lockdown (17 October 2020 to 31 January 2021). The first part of the study involved training an SVR model on the 75% of the UK dataset from wave 1 (n total = 435). This SVR model was then tested on the remaining 25% of data (MSE training = 2.04; MSE test = 2.29), which resulted in depressive symptoms to be the most important variable - followed by self-perceived loneliness. Statistical analysis of depressive symptoms by week of lockdown resulted in a significant U-shape pattern between week 3 to 7 of lockdown. In the second part of the study, data from wave 2 of the UK lockdown (n = 263) was used to conduct a graphical and statistical inspection of the week-by-week distribution of scores regarding self-perceived loneliness. Despite a graphical U-shaped pattern between week 3 and 9 of lockdown, levels of loneliness were not between weeks of lockdown. Consistent with past studies, study findings suggest that self-perceived loneliness and depressive symptoms may be two of the most relevant symptoms to address when imposing lockdown restrictions.

Social Media Posts Popularity Prediction During Long-Running Live Events A case study on Fashion Week

In the last few years, social media has dominated various aspects of people’s life including social events. Users participate more and more in long-running periodical events in social media, by sharing their experiences and preferences. This information provides unprecedented opportunities allowing businesses to promote their brands coverage by using word-of-mouth (WOM), that is enabled by the user generated contents (UGCs). Studying social media content popularity by considering the societies’ behavioral patterns is, therefore, paramount. In this thesis, we inspect users’ engagement motives in long-running events by means of a comprehensive statistical analysis of fashion week events on Instagram. Additionally, we develop a multi-modal approach to solve the problem of post popularity prediction that exploits potentially influential factors and apply it on fashion week events. We employ two metrics for implementing a filter feature selection technique, together with an automated grid search for optimizing hyper-parameters in four regression methods: ridge, support vector regressor, gradient tree boosting and neural networks.

An Efficient Method for No-Reference Video Quality Assessment

Methods for No-Reference Video Quality Assessment (NR-VQA) of consumer-produced video content are largely investigated due to the spread of databases containing videos affected by natural distortions. In this work, we design an effective and efficient method for NR-VQA. The proposed method exploits a novel sampling module capable of selecting a predetermined number of frames from the whole video sequence on which to base the quality assessment. It encodes both the quality attributes and semantic content of video frames using two lightweight Convolutional Neural Networks (CNNs). Then, it estimates the quality score of the entire video using a Support Vector Regressor (SVR). We compare the proposed method against several relevant state-of-the-art methods using four benchmark databases containing user generated videos (CVD2014, KoNViD-1k, LIVE-Qualcomm, and LIVE-VQC). The results show that the proposed method at a substantially lower computational cost predicts subjective video quality in line with the state of the art methods on individual databases and generalizes better than existing methods in cross-database setup.

Exploring the long-term changes in the Madden Julian Oscillation using machine learning

The Madden Julian Oscillation (MJO), the dominant subseasonal variability in the tropics, is widely represented using the Real-time Multivariate MJO (RMM) index. The index is limited to the satellite era (post-1974) as its calculation relies on satellite-based observations. Oliver and Thompson (J Clim 25:1996–2019, 2012) extended the RMM index for the twentieth century, employing a multilinear regression on the sea level pressure (SLP) from the NOAA twentieth century reanalysis. They obtained an 82.5% correspondence with the index in the satellite era. In this study, we show that the historical MJO index can be successfully reconstructed using machine learning techniques and improved upon. We obtain a significant improvement of up to 4%, using the support vector regressor (SVR) and convolutional neural network (CNN) methods on the same set of predictors used by Oliver and Thompson. Based on the improved RMM indices, we explore the long-term changes in the intensity, phase occurrences, and frequency of the winter MJO events during 1905–2015. We show an increasing trend in MJO intensity (22–27%) during this period. We also find a multidecadal change in MJO phase occurrence and periodicity corresponding to the Pacific Decadal Oscillation (PDO), while the role of anthropogenic warming cannot be ignored.