Prashanth Thirukkurungudi Sekar

The goal of this project was to identify undervalued soccer players in the top 5 European leagues.

1. Past 5 seasons of data from the top 5 European leagues was utilized to evaluate the market values of players. After feature engineering, multicollinearity was removed using variance inflation factor (VIF).

2. An ensemble of linear regression and gradient boosting was implemented to predict the market value and was hosted using a shiny dashboard. (Link)

The objective was to identify the relationship between BMI and Medical charges for people of different age groups and smoking habits

1. An exhaustive exploratory data analysis was performed to study the univariate and mutlivariate relationships present between BMI, Medical charges, age and smoking habits.

2. A local regression (LOESS) model was developed to capture the non-linear trend present in the data to explain the relationship present in the research question

In this project,the selected business questions were : Who were the revenue generating customers? ; How much revenue are they generating? ; What is their lifetime value to the business?

1. To identify the revenue generating customers, a combination of Logistic regression, random forest and gradient boosting classifiers were implemented resulting in a recall score of 0.885

2. A combination of Linear regression, random forest and gradient boosting regressors were implemented to forecast the revenue generated through the identified customers resulting in a RMSLE score of 1.625

3. For the monetary based ranking analysis, the Betageofitter and GammaFitter models were implemented

The objective of this project was to build a system to generate topics from the given corpus and further generate meaningful headlines depending on the input from the user.

1. Latent Dirichlet Allocation (LDA), Non-Negative Matrix Factorization (NMF) and Markov Chain models were implemented to generate topics

2. Long Short Term Memory (LSTM) networks was implemented to generate meaningful headlines depending on the input from the user

In this project, the requirement was to provide recommendations for items of one of the largest manufacturer in the US followed by identifying the probability of a customer returning in order to help their email marketing campaign.

1. After identifying the time frame for repeating customers and how to handle newly launched items, an Item-Based Collaborative Filtering algorithm with jaccard score as a metric was developed.

2. Since the returning rate of customers were too low, Synthetic Minority Over Sampling Technique (SMOTE) was implemented to balance the data followed by a logistic regression to predict the probability of a customer returning.

3. The results were implemented in a Shiny app and the A/B testing resulted in 2.5% of increase in sales.

The requirement was to project the sales of both newly launched and existing items for one of the leading manufacturer of furniture in the US by providing a month ahead and real time daily updates of the forecasts.

1. For the newly launched items, 12-month forecasts were provided using an ensemble of L2-Regularized regression and a fixed effects model by considering the trend and seasonality observed in items with similar characteristics such color, type, division etc.

2. For the existing items, after identifying new interactive features and basic transformations, a 3-month forecast with real time daily updates was provided using a L2-Regularized regression.

3. Power BI was used to represent and deliver the results

The task was to identify the importance and impact of the factors that play a vital role in the persistence of students into the following term of study

Given the binary nature of the target variable, a logistic regression was developed initially and it was converted into a marginal effects model in order to allow easier interpretation of the coefficients.

The results were represented through a waterfall chart for better visualization

The idea was to build an automated forecasting tool for providing generic forecasts for any given data set.
In order to provide stable forecasts with sufficient trade-off between bias and variance, several models were developed as follows:

1. Time Series Models: Several Time Series models such as Holt's Winter, ARIMA and TSLM were developed to provide forecasts
2. Regression and it's variants : Different variations in regression including lags, exponential weighted moving average, Lasso, Ridge and Elnet were developed
3. Machine Learning Models: Random Forest and Gradient Boosting was developed in addition to the above mentioned model
4. Non-Negative Least Squares Regression : A constrained regression was developed based on the above obtained results in order to compute the weighted average of the above developed models

The requirement was to identify the customers who would subscribe for a term deposit which would help the client in their marketing campaign.

After removal of outliers using the Box-plot and Cook's distance, PCA was executed to reduce the dimensionality in the data and K-means clustering was performed to segment customers and provide the clusters as an independent variable to the model.

After evaluating the individual performance of logistic regression and random forest with respect to F-Score, a weighted ensemble of the 2 models was taken to improve the precision and recall and a SHINY app was developed for real time interactions

PERL based API was used to extract and format the reviews from Amazon website. After basic cleansing of text by removing stop words, special characters, numbers, URL's etc, stemming and suffix dropping was performed.

The cleansing was followed by few interactive plots such as word cloud, word-frequency plot and Association plot.

Finally hierarchial and K-means clustering was performed to segment the reviews followed by a Naive-bayes classifier to estimate the sentiment of the reviews and implemented the solution using SHINY