AN ANALYTICAL VIEW OF KOLKATAS TEMPERATURE THROUGH MULTIVARIATE TIME SERIES MODELLING AND PREDICTION

In the course of development of a country like India, a huge challenge faced by peoples of India is irregular climate change. The recent increasing global temperature trend concerns over the seriousness and vulnerability of risk of lives of people of one Indian megacity Kolkata due to weather extremities caused by those irregularities in weather features. So, the prediction of Kolkata’s temperature in near future with more accuracy is a great challenge to climatologists, environmentalists and concerned policy makers. In this study, the time series of daily temperature records is decided as a function of weather predictors considered mainly, humidity, wind speed, pressure etc. The daily historical weather data from 1st March,2000 to 29th March,2024 has been taken with a dataset of daily basis as well as monthly basis averages of these weather variables. The data is trained and tested with various advanced time series algorithms like SARIMA selected by grid search mechanism in univariate model, LSTM, Prophet models as machine learning models in multivariate set-up. In the end, a new proposed extended multivariate prophet (EMP) based on principal components in a large dimension of multivariate set-up is executed as best considered by its forecast accuracy metrics. The proposed EMP is robust against multicollinearity problem and naturally free from overfitting. Based on the analyzation the temperature has been predicted for the future six months. The proposed model and results obtained in our study may prove to be very useful for policymakers concerned with weather related issues and environmental scientists.