Projects
Learning the Shifts in Choice Patterns for a Data Set Including Transactions of Food Products (Part of Thesis Research):
- Fitting and evaluating (using metrics like MRR, NDCG, $\chi^2$, KL) different ML models including Multinomial Logistic Regression and its extensions, Mixed Choice Models, and Markov-Based Models was done using sklearn and pyomo.
- Demand patterns by proposing and fitting new models incorporating context effects. Heterogeneity in customer types was explored using unsupervised learning models.
- Optimal Recommendations were obtained by solving a Conic Quadratic optimization problem with Gurobi solver under pyomo Gorubi.
Revenue per Click Optimization in Product Recommendation Modules in E-Commerce
- An ML model were designed which uses product’s features and click/impression history to rank the candidates in given recommendation models,
- Assortment Optimization Problem (combinatorial) was solved in order to obtain the best set set for recommendations with max expected revenue per click.
Finding Global Optima of the likelihood function of Multinomial Logit Model (Equivalently Multinomial Logistic Regression)
- Tuning and performance comparison of Gradient Descent, Coordinate Gradient Descent, Stochastic Gradient Descent, Barzilai-Borwein, and BFGS methods was done.
Near Optimal Hub-selection and Route Design; a Postal Company Case
- Mixed Integer Program was proposed to take into account timing, location, carrier constraints, and city demographics.
- Near-optimal solutions is provided by Linear relaxation to the problem.
Choice Pattern Recognition for Online Customers of a Hotel Chain
- Different Supervised Learning models were fitted to infer and predict the purchase behavior of customers’ room selection processes.
- The comparison made based on descriptive indices like AIC and BIC and Predictive Performance is measured by tests like the Chi-Square test.
- Demand sensitivity analysis to price changes is done and an algorithm for the optimal room price suggestion was proposed.
