Projects

Predicting the number of checked bags for a flight using Machine Learning

Baggage handling is an important part of the airport terminal service system. The baggage handling system gets overloaded or underloaded because of having no idea about how many checked bags a flight is going to have. This uncertainty hinders the timeliness and accuracy of the system for the lack of efficient manpower allocation.

To reduce uncertainty in handling checked bags, American Airlines provided 1 million flight information to develop a Machine Learning model which is capable of predicting the number of checked bags for a new flight. The dataset contains 60 variables that contain information about the passengers and the flight. Various Machine learning models such as Gaussian Process Boosting, Gradient Boosting, Support Vector Regression, etc. are adopted to find the best model for prediction. After model evaluation and parameter tuning, 80% prediction accuracy was achieved.

Identifying regional characteristics of transportation research with Transport Research International Documentation (TRID) data

This study explores the region-specific characteristics of transportation research based on large-scale bibliographic data collected from Transport Research International Documentation. Top geographic regions that are frequently used as study areas in transportation research are identified at both the country/region and U.S. state levels. An unsupervised machine learning model (latent Dirichlet allocation) was used to explore the region-specific characteristics of transportation with around 260k paper abstracts collected from TRID. The results confirm that transportation research exhibits clear region-specific characteristics, namely each geographic region is associated with a unique transportation topic distribution.

Second Harvest food bank delivery route efficiency study to maximize food delivery efficiency, Sustainability Fellowship, Fall 2020 (link)

Second Harvest provides healthy food and other needed items for low-income and unemployed people to relieve situations of emergency and distress. As the region’s primary source for charitable food, Second Harvest acts as a storage and distribution depot for over 149 smaller front-line agency partners in 17 Florida counties.

However, the cost of fuel and truck maintenance is expensive. The primary goal of this project is to critically assess current routes and provide recommendations for alternative routes that maximize delivery efficiency. After assessing the current practice, an efficient route planning optimization framework for the food delivery trucks to overcome inefficient manual planning so that the total operational costs are minimized.

Android App Development

Word Store: save, practice, and learn vocabulary (link)

This app is helpful to individuals who want to learn and practice new vocabulary, language or terms related to specific study fields. Using this app you can save new words with meaning, example sentences, mnemonics, synonyms, antonyms, notes etc. People who are learning new languages can also revise newly learned words using this app.

This app has been downloaded more than 100k times from Google Play Store and has a rating of 4.3 out of 5.

Paperless Time Study (link)

Time study is a method for establishing employee productivity standards in which a repetitive task is broken into simple steps, the sequence of movements taken by the employee in performing those steps is carefully observed, and precise time taken for each correct movement is measured. For manually conducting time study one needs a stop-watch, a study board and a time study form.

This app reduces the effort required to conduct a time study. You just need an Android smartphone instead of a stopwatch, study board or study form for conducting a time study. This app also generates a spreadsheet of all laps and foreign elements and shows a summary report of the study in a google doc file.

Wordler (link)

Wordle is a vocabulary game by NYTimes in which players get six tries to deduce a word of the day. A 5-wide by 6-long grid of empty white boxes is used where each box can hold one letter, and so you guess the answer by typing in a five-letter word.

Based on the rule of wordle, this app suggests the next probable words based on the previous inputs of a player. The suggested words are ranked based on the highest probability of guessing the next right word.