publications

2021

1. IEEE Xplore

   COVID-19 Detection Using Convolutional Neural Networks and InceptionV3

   Swarad Gat, Niharika Abhange, and Shilpa Paygude

   2021 2nd Global Conference for Advancement in Technology (GCAT), Oct 2021

   Abs HTML PDF

   The world today faces the grave crisis of a pandemic that threatens life, health, industries, economies and society as we know it. To tackle the spread of the COVID-19 disease, it is imperative that the individuals affected by it are identified efficiently. The medical workforce is overwhelmed in many countries, and doctors and nurses must be assisted with rapid automated systems that detect COVID-19 confidently without the need for trained personnel or special manufacturing processes. The currently available testing mechanisms are accurate but take up to 5 days to show results. Automated radiological testing is faster and does not require any testing kits. This study aims to simplify, speed up and validate the process of COVID-19 detection and differentiation from other lung diseases using Artificial Intelligence and Machine Learning principles to implement a cost effective and accurate COVID-19 test by detection of inflamed parts of lungs in chest X-Rays. Two approaches for separate target users are built. In the first approach, a customized Convolutional Neural Network model is built to classify chest X-ray images into two classes- 0(COVID-19 negative) and 1 (COVID-19 positive). This model reached an accuracy of 98%. The Inception v3 architecture (GoogLeNet) was trained for the second application, in which people infected by COVID-19 were distinguished from patients of other underlying lung diseases such as Atelectasis and Pneumonia. This approach demonstrates how reliably an automated COVID-19 test can be extended to people with other lung complications and showed a 91% accuracy.

2. IJERT

   Using Ensemble Method to Forecast Relative and Absolute Humidity

   Swarad Gat, Digvijay Patil, and Atharv Ganla

   International Journal Of Engineering Research & Technology (IJERT), Jun 2021

   Abs HTML PDF

   When it comes to the adverse influences of pollution, hurricanes, global warming, evolving weather patterns, and changing climatic conditions, humidity is an important feature of air quality. The main factors that come into play when deciding the humidity of a particular area are relative humidity (RH), which is the amount of water vapor present in air, and absolute humidity (AH), which is the amount of water found in a parcel of air. Since precise humidity prediction is important, the aim of this study is to use Machine Learning and Deep Learning techniques to predict the values of Relative and Absolute Humidity. The RH and AH values are predicted using two simple algorithms. The values are also combined, and the final figures are forecasted and analyzed. In the problem of forecasting relative and absolute humidity, the findings demonstrate the utility of the ensemble model, which outperforms the use of single model prediction.

2022

1. Springer

   Hip-Hop Culture Incites Criminal Behavior: A Deep Learning Study

   Swarad Gat, Niharika Abhange, Rahul Jadhav, and 3 more authors

   Intelligent Sustainable Systems, Jan 2022

   Abs HTML PDF

   For centuries, music has been an inseparable part of many human cultures. The rise of the hip-hop culture over the last 50 years has turned into a powerful movement, empowering people from various communities and making their voices heard. However, certain parts of hip-hop and rap music have started being associated with misogyny, substance abuse and violent behavior. This study aims to find a correlation between lyrics of hip-hop and rap songs that glorify such illicit behavior through their lyrics and the actual rate of criminal activity of individuals that are directly or indirectly influenced by hip-hop culture. This research employs NLP concepts to build a model that detects song lyrics that falls into any of the 3 categories—“Misogyny,” “Substance Abuse” and “Violence.” A comparative study is conducted by training multiple models including multinomial naïve Bayes, random forest and LSTM on a manually collected and labeled dataset consisting of rap song lyrics released between 1970 and 2020. The highest performing model (LSTM—87% accuracy) was subsequently used to detect objectionable lyrics in popular rap songs of the decade of 2010–2019. To obtain a correlation of these with the criminal activity of the target population, official data of criminal activity (2010–2019) of citizens aged 0–29 from the largest hip-hop influenced areas in the world are compiled. This dataset is analyzed and to obtain strong evidence of a correlation between objectionable rap song lyrics being promoted through song lyrics and the criminal tendencies of the youth that is primarily affected by it.