Doctoral Research

Problem domain: Climate Data Mining for Indian Monsoon

Abstract: Indian summer monsoon rainfall (ISMR) is a dynamic process, which considerably affects the economy of the country. It results from multiple sea-atmosphere interactions throughout the globe. Predicting ISMR is a challenging task due to presence of multiple influencing climatic parameters and involvement of complex mechanism.

Identification of climate indices influencing Indian monsoon and modelling Indian monsoon process is prime focus of the work. Different statistical and machine learning approaches are considered for discovery of novel climatic indices important for monsoon process and designing the forecast models. An attempt is made to study the variability of Indian monsoon over temporal window. A joint clustering approach of years and predictors is adopted to cluster the years according to different subsets of climatic parameters and homogeneity of patterns, thereby building the forecasting model with improved precision.

Climate index discovery assists in visualizing different aspects of climatic system. We focus on discovery of climate indices important for ISMR from climatic variables, namely, surface pressure, sea surface temperature, and wind velocity. Community detection approach is applied for the purpose. Discovered climate indices are found to be highly correlated to Indian monsoon process and they also ascertain their superiority in prediction of ISMR.

Deep Learning-based architecture is also utilized to discover novel climate features that are relevant to Indian monsoon. It attempts to model high-level abstractions in data by using architectures composed of multiple non-linear transformations, where features are discovered and improvised at every layer. Finally, we aim to design a hybrid model by combining physics based global climate model and machine learning based model. Thereby, encashing the benefits of both models: (i) Physics based computational models move from knowledge (in the form of physical theories of climatic processes) towards data - a way of exploring how well current theory explains the data, and (ii) Machine learning based approach moves from data to knowledge - a mathematical model that describes the discovered relationships among the data.

Supervisor: Dr. Pabitra Mitra, Department of Computer Science and Engineering , IIT Kharagpur.

Research in Master of Engineering (ME)

Thesis title: Fuzzy Rough Set Approach to Develop an Adaptive System

Abstract:An adaptive system is a set of interacting or interdependent entities, real or abstract, forming an integrated whole, and able to respond to environmental changes or changes in the interacting parts. Development of an adaptive system is accomplished by three main steps. First, dimension reduction is performed utilizing fuzzy-rough concept and genetic algorithm. Secondly, optimized range of values to be associated with each linguistic label of the attributes are evaluated with the aid of hierarchical Q-learning. Finally, an efficient classifier is designed, which is preceded by evaluation of most important fuzzy rules (Rule Mining) using the concept of multi objective genetic algorithm.

Supervisor: Prof. Jaya Sil, Department of Computer Science and Technology, IIEST Shibpur.