Hi!

Welcome to my growing collection of projects related to data science, machine learning and NLP.

---

Ingredient detection and recipe recommender

The key aim of this project was for households to utilise existing ingredients and whip up a meal, with a vision of expanding into smart home appliances and/or end-to-end user shopping experience. In 2 weeks, I custom trained a convolutional neural network model (Yolov8m) to detect ingredients from images and suggest Singaporean recipes. I achieved 75% mean average precision across 30 ingredient classes that are common to local recipes. The demo was eventually deployed via FLASK at Pythonanywhere.

Fig (left). example of predicted ingredients

Fig (right). mean average precision(IoU 0.5) of 0.75 across 30 classes

[View code at Github] [Demo app at Pythonanywhere]

Tools used: Python, Jupyter, NLP, Roboflow, Google Colab, FLASK

Category: Computer vision, Deep Learning

Year: Aug 2023 (ongoing)

---

Predicting virus in mosquitoes

The key aim of this project was to predict virus presence in mosquitoes for the city’s ops and resource planning in containing the vector-borne virus. I built a gradient boosted decision tree model, achieving 88% ROC AUC. It was found that weather factors related to “month”, “temperature”, “time of sunset” were strong predictors of virus presence.

Fig (left). example of predicted risk in a map

Fig (right). confusion matrix depicting high recall

[View code at Github] [Demo app at Streamlit]

Tools used: Python, Jupyter

Category: Predictive modelling, Decision trees, Classification, Kaggle Challenge

Year: July 2023

---

Classifying specialty diet preference using NLP

The key aim of this project was to classify and predict a user’s specialty diet preference based on text inputs. Subreddits for Keto and Paleo diets were scraped, insights extracted, and features extracted for creation of the eventual logistic regression model (93% F1 score). It was found that words like “fat” and “cream” tended to predict Keto over Paleo.

Fig (top). Exploratory analysis of both subreddits

Fig (bottom). SHAP feature importance plot

[View code at Github] [Demo app at Streamlit]

Tools used: Python, Jupyter, NLP

Category: Predictive modelling, Logistic Regression, Classification

Year: June 2023

---

Predicting HDB resale prices

The key aim of this project was to predict Singapore’s HDB resale prices based on historical datasets. A ridge regression model was developed with a 90% accuracy score based on 34 features. A bigger floor area, higher floor and hawker centres nearby were found to be strong positive predictors. Contrary to popular belief, distance to a good primary school was not found to be a strong predictor.

Fig (top). feature importance plot

Fig (bottom). plot of predicted versus actual values

[View code at Github] [Demo app at Streamlit]

Tools used: Python, Jupyter

Category: Predictive modelling, Linear Regression, Kaggle Challenge

Year: May 2023

---