Hello!!!

I'm Sai Sree

I am a dedicated and innovative Machine Learning Engineer/Data Scientist. My passion lies in leveraging data-driven insights and advanced machine learning algorithms to solve complex real-world problems.

LinkedIn Profile

Projects

In-Progress

Prediction of Optimal Income Distribution and Savings using ML

Skills: Machine Learning, Python
Prediction is based on:

  • Distribution of income into desired percentages
  • Allocation of savings amount for fixed tenure

Will be completed soon
In-Progress

Prediction of Number of Electrical Devices in Commercial Buildings

Skills: Machine Learning, Python
Prediction is based on:

  • Number of employees working
  • Square-feet of building and number of rooms
  • Optimizing power bill and thus predicting devices

Will be completed soon
In-Progress

Estimating the Effect of a Member Rewards Program using DoWhy and EconML

Skills: Python, Causal Inference
Estimation is based on:


Will be completed soon

A/B Testing on Digital Marketing Dataset

Skills: A/B Testing, Hypothesis Testing, Statistical tests, Python
Test was done by,

  • Using null hypothesis and alternative hypothesis
  • Using Chi-Square test for quantitative variables
  • Using Levene test for qualitative variables
  • Using t-test

Test result was concluded that the customers exposed to advertisements tend to convertion compared to the customers who were not exposed to them.

Estimating the Effect of a Member Rewards Program using DoWhy

Skills: Causal Inference, Python, DoWhy
Estimation is done based on:

  • Whether the rewards are effecting the member or not

Results revealed that the members who are rewarded has greater impact compared to the members who were not.

Deep Learning for the Classification of Pseudogenes in the Genome

Skills: CNN, LSTMs, Bi-directional GRUs, Model Optimization, Python
Classification is made in between,

  • Coding regions (from NCBI)
  • Non-Coding regions (from RNA Central) and
  • Pseudogenes (from pseudogene.org)
  • CNNs, LSTM, Bi-directional GRUs, and Hybrid model in which CNN and Bi-directional GRU were combined for better classification.

Model was evaluated based on different criteria namely training time, accuracy, precision, cohen's kappa and NIR

Stock Market Price Prediction using Deep Learning Techniques

Skills: RNN, LSTMs, Model Optimization, Python
Prediction was made:

  • On CAC40 stock market data (source: Kaggle)
  • Data was pre-processed and scaled using Min-Max Scaler
  • Using LSTM and LSTM-GRU models
  • Various hyperparameter tuning techniques were implemented to reduce the loss
  • Tuning has played an important role in the implementation of model as the loss has reduced up to 20%.

The best model is the model with the lowest loss, and it is used for prediction of values for test data. The best model was with 32 neurons, batch size of 8, and 0.2 drop out values.It is difficult to build and try the model for every given value of number of neurons, batch size, activation function, optimization, drop out values, the tuning has been made as an iterative process, to automatically select the best parameters for the model.

Prediction of Customer Churn for an Internet Service provider

Skills: Tableau, EDA, Decision Trees, Random Forest, and Clustering Methods
Customer Churn prediction is

  • conducted on the dataset with 72274 observations and 11 attributes.
  • Statistical analysis of variables
  • made using decision trees, random forest, gradient boosting
  • customers were grouped using clustering technique in order to provide particular solution for each group

Descriptive analysis was made using Tableau, exploring the dataset with different possibilities, and conclusions were drawn. The next objective refers to classification of churner and non-churner customers. This prediction was achieved by implementing supervised models namely decision trees and gradient boosting. Gradient boosting with lowest misclassification rate has been considered as the best model. In addition to that, unsupervised clustering has also been implemented to cluster group of customers into segments.

Contact Information