My first Kaggle competition - Women in Data Science (WiDS) 2021

As a beginner to the Data Science and ML world, I had no idea how these competitions will be. I always thought such competitions were for the experts. And then I happened to attend a session from Kaggle Expert Usha who decoded many aspects of a competition in Kaggle and motivated me and others to take part in the competition.

The competition was "Women in Data Science (WiDS) Datathon 2021". This competition's purpose was to inspire women to learn about data science and also to create a supportive environment. This competition was open to males as well, however at least 50% of the members in the team should be women. So I partnered with my friend and ex-colleague.

This year's competition focused on models to determine whether a patient admitted to an ICU has been diagnosed with a particular type of diabetes, Diabetes Mellitus. Usha had shared a notebook with a lot of background work done for this problem. So her notebook became a bible to understand the whole process.

I always heard from people that feature engineering is a major task in an ML project and I experienced myself why people say so. 

The training data and test data were shared as part of the competition. There were 180 features and a label. The training data had 130157 rows with label and the test data had 10234 rows without label, of course.

As part of the feature engineering, the first step was to understand these features. Without a domain knowledge it was not an easy task to understand how many of these features are useful or not. As a first step, I had to understand both numerical and categorical data. Then removed a few columns which I thought do not add value to the model like hospital room number. Then removed a few columns which had more than 70% missing values. After every step I was checking the accuracy to understand the changes it brings to the model. As suggested by experts, we used CatBoost algorithm. CatBoost is an open source algorithm and is based on gradient boosted decision trees. With some trial and error, we got an accuracy of 83.941% and we were in top 64% in Leaderboard.

Keeping aside the position in Leaderboard, key takeaway was understanding how to solve a business problem using ML. I got to know how to apply feature engineering and exploratory data analysis. Due to other commitments, I could not spend more time on this competition but whatever I spent I got a good experience on a ML project. I'm excited to compete in more competitions and that will only improve my knowledge and experience in ML.   

Comments

Post a Comment

Popular posts from this blog

Ordinal Encoder, OneHotEncoder and LabelBinarizer in Python

Usually in the data set we work on, we will have both numerical data as well as categorical data (non-numeric). There are algorithms which do not work on categorical data. These algorithms are designed that way to enhance efficiency. So in such cases how do we train the model with categorical data? This is where we need to apply a technique called Encoding. Encoding transforms categorical data into numeric. There are various Encoders available in sci-kit learn library. In this post, I'll share my learning on three Encoders. Ordinal Encoder This performs ordinal (integer) encoding of categorical data. Let's look at an example. I have an array of categorical data. This is a 2 dimensional array. multiArray = np.array([[ 'Karnataka' ,  'KA' ], [ 'Maharastra' ,  'MH' ], [ 'Gujarat' ,  'GJ' ]]) Then I use OrdinalEncoder to transform this data. ordinalEncoder = OrdinalEncoder() ordinalEncoderArray = ordinalEncoder.fit_transform(mult...
Read more

Data Visualization using Pandas - Univariate Plots

Image
Data visualization is an important step in ML. That helps us to understand the data in visual representation. Why does it matter? Because it helps us to understand the distribution of the data, any outliers, and also relationship between multiple attributes. Experts also mention that understanding the distribution helps us in choosing correct algorithm. I cannot comment on this statement as of now, as I do not have enough knowledge on that. When I reach that stage, I’ll definitely share. Pandas library provides methods to plot data in multiple ways. In fact, I did use one of those plots called  Box and Whisker Plots  in my previous examples. However, those plots were used to see the difference between multiple algorithms. In this post, I’ll share how to visualize the data which is an important step much before we apply any ML algorithm. I’ll use the data set which was used for  Binary Classification . The data set is   Pima Indians Diabetes Database  which ...
Read more

Stemming in Python

While working on NLP, we come across huge volume of words. Before using these words to the ML algorithm, a lot of pre-processing happens. These pre-processing ensures we keep relevant words which can then be used to train an ML algorithm. One such technique is called Stemming . As defined in Wikipedia - "In linguistics, a stem is a part of a word used with slightly different meanings and would depend on the morphology of the language in question." A simple example is the words approximation , approximate , approximated and approximately have the same root. So as part of Stemming, we need to make sure these words should be taken as a single word instead of four different words. So how do we achieve this in Python? The NLTK library provides support for Stemming. There are two algorithms for stemming - PorterStemmer and LancasterStemmer . I'll not get into the details of how these two works, but at a high level LancasterStemmer does a lot more iterations compared to Porte...
Read more