KSI ML Collison Fatalness Predictor

Predict whether a Toronto traffic collision will result in a fatal or non-fatal outcome using supervised machine learning, served through a Flask REST API and a Next.js web frontend.

Available to predict over 5 models with highest accuracy of 88%

Trained over 18,000 records or real data from Toronto Police Depart.

Implemented clean data preprocessing using NumPy, Pandas and Seaborn for important feature selection and data cleaning

Using SMOTE synthetic technique in order to balance classes in training data.

Combined highly presence of categorical values into common groups in order to ease the process of OneHotEncoding for numeric conversion.

Implemented chi2 square similarity for finding relations between target feature and other features.

Trained an SVM model on the training set and fine tuned the model using GridSearchCV, with achieving an accuracy of 83.6% after finetuning.