http://dspace.dtu.ac.in:8080/jspui/handle/123456789/26 2026-07-22T07:48:35Z 2026-07-22T07:48:35Z CHAUDHARY, PRATHA YAKSHITA Gupta, Anjana (SUPERVISOR) http://dspace.dtu.ac.in:8080/jspui/handle/repository/23007 2026-07-06T09:16:09Z 2026-05-01T00:00:00Z

Title: YAKSHITA PRATHA MODEL (YPM ): AN ALTERNATIVE TO DUCKWORTH LEWIS STERN (DLS) METHOD Authors: CHAUDHARY, PRATHA; YAKSHITA; Gupta, Anjana (SUPERVISOR) Abstract: The Yakshita Pratha model (YPM) is an innovative machine learning model designed to set accurate, revised target scores for limited over cricket matches that are interrupted by inclement weather, floodlight failure or some crowd issues. While the former Duckworth Lewis- Stern (DLS) method operates on static mathematical framework centered primar ily on resource preservation (wickets left and overs remaining ), the YPM captures modern cricketing dynamics . This model is grounded in a series of robusts data-driven assump tions regarding a team’s evolving behavioural patterns, scoring potentials and a highly non-linear impact of losing wickets as the match progresses. The core philosophy of this research is the transition away from the static models to wards a highly adaptive , observe scoring paradigm. In contrast to the traditional method that rely heavily on generalized baseline decay curves, the YP model is based on mod ern shifts in playing style . It accounts for contemporary strategic changes, including deep batting lineup, aggressive data driven strategies and multi phased power hitting . This model treats match scoring as interconnected, path dependent process, the proposed model conditions, predictions based on how a team has used their resources in previous matches and real time execution context , rather than theoretical remaining capacity.

2026-05-01T00:00:00Z SAINI, SHREYA YADAV, TANVI Gupta, Anjana (SUPERVISOR) http://dspace.dtu.ac.in:8080/jspui/handle/repository/22898 2026-06-25T04:53:29Z 2026-05-01T00:00:00Z

Title: PREDICTING CROP YIELDS USING MACHINE LEARNINGWITH SHAP-BASEDEXPLAINABILITY Authors: SAINI, SHREYA; YADAV, TANVI; Gupta, Anjana (SUPERVISOR) Abstract: The agriculture sector is one of the pillars of food security and development world wide. Accurate estimates of crop production are valuable for making decisions in the field of resource planning, supply chain management and policy making. However, con ventional methods of yield prediction, relying on past patterns or basic statistical models, come with limitations in being able to account for the nonlinear relationships found in agriculture. Predicting yield based on environmental, meteorological and agronomic data is a machine learning approach in this project. This data is taken from the Food and Agri culture Organization (FAO) and is available on Kaggle, which consists of 28,242 agricul tural records from 130 countries from 1990– 2013. Five different model regression were developed and used for testing: Linear Regression, Lasso Regression, Ridge regression, K-Nearest Neighbours (KNN) and Decision Tree Regressor. The input variables with temporal (year), climatic (average rainfall, temperature), agronomic (pesticide use) and categorical variables (crop type and geographic region). Handling missing values, one hot encoding of categorical features, and feature standardization using a scikit-learn’s ColumnTransformer pipeline comprised the preprocessing steps. Results have shown that DT Reg performed best on generalization with an R² value of 0.9793 and Mean Squared Error (MSE) of 3941, an improvement of 23% over the baseline LR model (R² = 0.7473). More interpretability was achieved by using the SHapley Additive exPlanations (SHAP) analysis: crop type, especially potatoes, turned out to be the most important and dom inating factor, while pesticide use and climatic factors followed. These nonlinear rela tionships were identified using feature-level analysis via SHAP dependence plots, with pesticide use showing a strong negative correlation with yield for high productivity sce narios, and temperature and rainfall having complex interactions, which varied by crop type and region. The project is finalized with the deployment-ready predictive system, and the discussion of practical applications in precision agriculture.

2026-05-01T00:00:00Z VANITA Kumar, Dhirendra (SUPERVISOR) http://dspace.dtu.ac.in:8080/jspui/handle/repository/22897 2026-06-25T04:53:22Z 2026-05-01T00:00:00Z

Title: AN ENHANCED DIVERGENCE-BASED DISTANCE MEASURE FOR INTUITIONISTIC FUZZY SETS WITH HESITATION INFORMATIONAND ITS EXTENSIONS TO INTERVAL-VALUEDAND PICTURE FUZZYENVIRONMENTS Authors: VANITA; Kumar, Dhirendra (SUPERVISOR) Abstract: Distance measures for intuitionistic fuzzy sets (IFSs) are central tools for pattern recognition, clustering, and multi-attribute decision making (MADM) under uncertainty. Although numerous divergence-based and geometric distance mea sures have been proposed in the literature, most of them either neglect the hesita tion degree, which carries genuine epistemic information, or fail to discriminate between sets that exhibit equal membership-non-membership differences but distinct hesitation profiles. To overcome these limitations, this paper proposes a novel enhanced divergence-based distance measure D+ L for IFSs that incor porates an explicit hesitation term derived from a modified Kullback–Leibler divergence. The measure is constructed from a single, symmetric core function and is shown to satisfy all four axiomatic requirements of an IFS distance metric, namely boundedness, separability, symmetry, and monotonicity. The proposed measure is further extended to two important generalizations of intuitionistic fuzzy theory: a six-term version DIV+ L for interval-valued intuitionistic fuzzy sets (IVIFSs) that fully exploits both the lower and upper bounds of member ship, non-membership and hesitation intervals, and a three-component version DP L for picture fuzzy sets (PFSs) that handles positive, neutral, and negative memberships. Six classical benchmark cases and two innovation-management decision problems are recomputed entirely from scratch using the proposed measure. Comparative analysis with twelve existing measures shows that D+ L resolves the counter-intuitive ties that plague competing measures, distinguishes hesitation-sensitive cases that earlier divergence-based measures could not, and yields stable rankings in TOPSIS-based MADM. The results confirm that the enhanced measure is mathematically rigorous, computationally concise, and practically effective.

2026-05-01T00:00:00Z RAJAN GUPTA, TRASHA (SUPERVISOR) http://dspace.dtu.ac.in:8080/jspui/handle/repository/22896 2026-06-25T04:53:11Z 2026-05-01T00:00:00Z

Title: LAND USE AND LAND COVER CLASSIFICATIONUSING SATELLITE IMAGERY AND DEEP LEARNING Authors: RAJAN; GUPTA, TRASHA (SUPERVISOR) Abstract: The precise and effective monitoring of land use and land cover (LULC) is important for urban planning, environmental protection, agriculture, and disaster management. Current approaches for classifying satellite images are dependent on the manual inspection process, which is time-consuming and can cause errors. Hence, a deep learning and computer vision-based solution for efficient LULC classification has been presented in this dissertation. For the purpose of developing our model, we used the EuroSat data set. It is a publicly available data set generated from the images collected by the Sentinel-2 satellite, which is owned and operated by the European Space Agency. Our deep convolutional neural network (CNN), inspired by the LeNet-5 architecture, is intended to classify satellite images of size 64x64 into ten classes such as Annual crop, Forest, Highway, Residential area, and Water bodies. For implementation, we used PyTorch, the popular deep learning framework, in conjunction with a GPU and CUDA. As a part of this dissertation, we have also considered the use of transfer learning. For that, we froze all but the last layer of a ResNet-18 CNN trained on the Imagenet database. It was shown through experiments that the proposed lightweight custom CNN model was able to achieve a validation accuracy of 66.9% in just five training iterations while taking only 2.44 minutes to train on the GPU altogether. The findings show that although deep learning models present themselves as a more reliable option compared to manual classification techniques, the similarity of spectra among vegetation classes cannot be ignored as the source of any possible mistakes. The findings suggest that small CNN architectures are an excellent choice for achieving real-time environmental monitoring in a more effective manner, and transfer learning with ResNet-18 represents a way forward to obtaining better quality. In total, these two concepts present a great platform for making further improvements using the multispectral analysis method and pixel-wise semantic segmentation techniques. (Keywords: Land Use and Land Cover (LULC), Convolutional Neural Network (CNN), EuroSAT, Transfer Learning, ResNet-18, PyTorch, Remote Sensing, Sentinel-2, Deep Learning, Image Classification.)

2026-05-01T00:00:00Z