One of the problem statements in automated insulin injetion for Type 1 Diabeties treament is determining the quantity and timing for ingestion of the insulin. In this exercise, we contribute towards prediction of timing of ingestion by guessing meal intake from Continous Glucose Monitor's data.

Given multiple timeseries from CGM data for a 2.5 hour window, our objective is to predict wheather the timeseries belongs to a meal intake or not.

Models used with accuracies

1. Logistic Regression (91%)
2. Random Forest (82%)
3. GaussianNB (67%)
4. Support Vector Machines (54%)
5. K-Nearest Neighbor (61%)

Train Accuracies: Test Accuracies:

Salient features:

• Data clean-up (Handle columns with >50% NaN values)
• Pre-processing (Combination of Forward fill, Backward fill and Range Fill)
• Grid Search for Hyperparameter Tuning
• Principal Component Analysis for Dimensionality Reduction; suggest top features
• Save and load model pickle file with model and weights
• Modular application - seperation of concern

1. Make sure scikit-learn, pandas, numpy and pickle are installed in your environment.
2. Change to repo directory
3. Make sure the test data doesn't have Empty rows or unequal lengths of rows. If there are unequal rows, pad them with NaN
4. Run train.py

python train.py

This will save weights in .sav files in working directory 5. Run test.py

python test.py

6. If you train data with PCA, enter 'YES'/1 when prompted for PCA in test.py

Test Data format:

• 1 csv file with similar structure as that of train data. Refer to Train Data.
• 1 ground truth file in csv. Use 1 for Meal and 0 for Nomeal in one-column format.

Note: While rows with >50% NaNs have been excluded from training, avoid passing such inputs

• Accuracies of models on training data
• Scores from 10-fold Cross Validation for each model
• <input> Test File Name (present in same working directory)
• <input> Choose to reduce dimensions of test data (PCA: Yes or No)
• <input> Choose Model

All models were randomly picked by team members for implementation