- Built a password generator
- Created a hangman game using python
- Created an auction program
- Created a blackjack game
- Built a number guessing game
- Build a quiz game
- Built a snake game using turtle graphics
- Built a pong game using turtle graphics
- Built a crossy road game using turtle graphics
Data Science and Machine Learning Projects
- Built a Streamlit based probability distribution fitter web application and deployed it in Heroku cloud. The application can be used to compare 80+ probability distributions and to identify the distribution that best fits the data. [Blog; Web App]
- Built a Streamlit based early diabetes prediction web application and deployed it in Heroku cloud. The application comprised of an exploratory data analysis section (summary statistics and plotly based interactive plots) and a machine learning-based diabetes prediction section. The ML section provides class probabilities and prediction based on inputs. [Web App]
- Simulated email link click-through rate using Bayesian A/B testing. In this project, two variant of email (with and without image header) were sent to two random groups. A weak prior probability is assumed and real user click data was gathered. Next, a Monte Carlo simulation was performed to identify which variant of the email is performing better and by what extent.
- Created sales and HR analytics dashboard using Tableau and MySQL. [Sales: Dash 1; HR: Dash 2]
- Built a parking space counter using OpenCV library. The application counts the number of empty and occupied space from a video feed of parking space.
- Segmented customers based on their purchase patterns using different unsupervised clustering techniques (K-means and Hierarchical).
- Predicted employee churn using Cox Proportional Hazard model (survival analysis: Python’s lifelines library). The model could be used to predict when an existing employee would leave an organisation. The advanced identification could be used to understand the existing problems and propose some remedies to retain employees for long term.
- Predicted lung cancer patents’ survival using Cox Proportional Hazard model (survival analysis). [Blog 1; Blog 2]
- Forecasted global temperature increase based on average annual temperature data collected from Central Park, NY using ARIMA time series model (the data was downloaded from the NOAA: 1870-2016).
- Built a diabetes prediction classification model explainer using LIME. The LIME is utilised as it approximates black box machine learning model with a local, interpretable model to explain each individual prediction. [Blog]
- Built a concrete strength prediction regression model explainer using Shapley values. The Shapley value method is utilised to identify the contribution of each variable in the model and its direction of influence (+ve or -ve). [Blog]
- Built a concrete strength prediction model (gradient boosted trees) using sklearn library and tuned its hyper-parameters using Grid, Random and Genetic-based search in Python). [Blog]
- Optimised deep learning CNN model hyper-parameter using Keras and sklearn. [Blog]
- Built a CNN-based MNIST data classification model using Keras (TensorFlow) and PyTorch [achieved 98.97% classification test accuracy]. [Blog]
- Built a rock paper scissors CNN classification model using Keras.
- Formulated plan and conducted reconnaissance survey across Kolkata city for Ph.D. data collection. Further, collaborated with eight students and collected video graphic data of pedestrians’ road crossing behaviour from 11 intersections across Kolkata city.
- Designed distraction themed questionnaire, estimated required sample size, trained interviewers for conversation styled interview and conducted face-to-face interviews across Kolkata city.
- Analysed the parents’ role in school mode choice for their children in Guwahati city using multinominal-logit model. The study highlights the role of parents in the mode choice process through their perceptions of safety, economic standards, and child characteristics.
- Analysed pedestrian foot-over bridge utilisation across four cities (14 locations) using tree-based ensemble techniques.
- Identified social and non-social factors influencing pedestrian’s signal violation behaviour at intersection crosswalks using binary logistic regression model.
- Identified factors influencing pedestrian’s distracted road crossing behaviour at signalized intersection crosswalks.
- The study identified the unsafe behaviour displayed by the distracted pedestrian during road crossing at signalised intersection crosswalks.
- Identified the optimal waiting time of pedestrian at intersection crosswalks using survival analysis (COX Proportional Hazard and Accelerated Failure Time models). The study results could be used to propose optimal red-phase length for pedestrian signal’s at intersection crosswalks.
Evaluated the qualitative factors that diminished or increased system usage for both bus and train users.
Designed rainwater harvesting system for engineering college.
Designed Higher and Middle Income Group Community Residential Buildings.