Gut Microbe Disease Classification

• 

  Landing page

• 

  Distribution of bacteria

• 

  Gut Disease Classification labels

Inspiration

The human gut is teaming with millions of microbes. Based on current research these microbes can have a significant impact on human health. Being able to understand the bacteria and it link to potential disease can help doctors in potentially catching diseases early.

What it does

This machine learning model helps classify gut health based on about 1094 different bacteria. During exploratory analysis it showed that the data was sparse, multi-class and imbalanced. For classification models, accuracy would be the key metric but due to the data being imbalanced and multi-class f1-score and cohen kappa were used.

How we built it

It was built using python and done using google co-lab. Data exploration was first performed to understand the different aspects of the dataset. It should it was a multi-classification problem. It also showed there was a data imbalance. Since the data had so many potential predictors balancedRandomclassifier was used to select key features. Imblearn library was used to tackle the data imbalance. The data was then split into train and test and then multiple machine learning models were applied. Mostly tree based model since the data was quite sparse. A stacked ensemble model was also used and this gave the best result.

Challenges we ran into

The main challenge was testing and experimenting with different machine learning models.

Accomplishments that we're proud of

Was able to leverage an ML model to potentially help doctors as a diagnostic/screening tool. Also happy that key measures for the project f1-score and cohen kappa were 93.6% and 91.5% respectively.

What we learned

Doing exploratory analysis is key as it helped uncover the data sparsity and imbalance. This helped in shaping which ML model to start of with.

What's next for Gut Microbe Disease Classification

In future the classification model could be incorporated in a web app which health practitioners can use to predict potential diseases in a patient. Could potentially be used as a diagnostic tool but it would need more rigorous testing.

Built With

• google-colab
• python