Hello, my name is Kevin Nduati

I'm a Data Scientist / Data Engineer

I am a talented Data Scientist equipped with mathematical, statistical and technical know-how to tackle complex issues and the ability to pursue solutions through advanced data analysis. I'm also hardworking, ethical and detail-oriented.

Find me on:

About Me

I'm a Data Scientist / Data Engineer

I have extensive knowledge in building and deploying data-intensive applications, overcoming complex architectural, and scalability issues in a variety of problems. I am proficient in data mining, data visualization, data processing, predictive modeling, as well as scripting languages, including Python and Scala. Capable of creating, developing, testing, and deploying highly adaptive diverse services to translate business and functional qualifications into substantial deliverables. I am also open-minded to learning new tools, and capable of working with other people.

Available: Yes

Degree: Economics and Statistics

Email: nduatikevin1@gmail.com

Phone: +254721679735

City: Nairobi

Download CV Enquiries
Python
75%
HTML
60%
CSS
60%
Scala
15%

Experience

08/2022 - Present

Junior Data Scientist

Xetova

  • Contribute to the development of next generation business analytics using predictive modeling, optimization, simulation and data visualization.
  • Contribute the the development of APIs and the deployment of machine learning models
  • Support the delivery of data solutions through automation, high quality visualizations and analytics
  • Building dashboards that assist in the supply chain process

06/2021 - 03/2022

Data Entry Intern

General Office Technology Solutions

  • As a data entry intern at General Office Technology Solutions, I worked on building a data entry system to cater to the 400+ clients.
  • The aim of the system was to enable a seamless migration from an analogue system that involved tracking payments using physical ledgers, to a digital system using Microsoft Excel.
  • We developed a system using Microsoft Excel, which reduced the time spent on manual work by over 50%. It also made easier the tracking of unpaid bills and arrears.

Education

2018-2022

Bachelors of Economics and statistics

School: University of Nairobi

Grade: Second Class Honors Upper Division

Majored in statisics and excelled in units such as:

  • Basic Algebra and Mathematics
  • Hypothesis Testing
  • Probability and statistics
  • Time Series Analysis

    • 2013-2016

    Kenya Certificate of Secondary Education

    School: Murang'a High School

    Grade: A-

    Excelled in Mathematics and Chemistry. I represented the school in Mathematics contests, and was the best in several of them. I was also a member of St Johns Ambulance where I learnt first aid.

    Tools and Platforms

    I am conversant with :

    Projects

    View on Github

    Sarcasm Detection

    • I tackled a text classification problem on sarcasm detection by comparing two state of the art models. I first cleaned the data using regex to remove numbers, spaces and punctuation.
    • The first model used word embeddings generated by Word2Vec. The embeddings were then tokenized and padded in order to fit to a bidirectional RNN. This model generated an accuracy of 78%.
    • The second model used "Bert-based-uncased" transformers from the Hugging Face library. This model not only trained faster, but also had a higher accuracy of 93%.

    View on Github

    Market Analysis using Clustering

    • A mall had customer data and wanted to group its customers so as to send personalized ads. I fit a simple Linear Regression model and checked the feature importances to get a better understanding of the features.
    • I also carried out dimensionality reduction using PCA and T-SNE to check for patterns in the data.
    • Carried out K-Means Clustering where the number of clusters was determined using Silhouette Visualizer. 5 clusters were deemed optimal
    • I then grouped the customers in the different clusters in order to improve ad targeting depending on spending and levels of income.

    View on Github

    Bank Customer Churn Modeling

    • Performed data mining and data visualisation techniques on the data to draw insights from the data.
    • I performed data balancing techniques by comparing undersampling techniques such as simple undersampling, nearmiss technique with oversampling techniques such as simple oversampling, SMOTE and SMOTE Tomek Links.
    • I chose to fit an XGBoost model on SMOTE balanced data, since it can scale up to a much larger size if the sample size is increased. ROC AUC score increase to 74%.

    View on Github

    Churn Detection

    • The data contains bank customer data, and I try to predict whether a customer subscribed a term deposit. I first created a python script that contained a Logistic Regression model fit on the data
    • I used hyperdrive from Azure Machine Learning Studio to tune hyperparameters(C and max_iter). I chose a random parameter sampler, and used bandit policy for early stopping. This run yielded an accuracy of 91.4%
    • I then run AutoMl on the same data, and managed to get an accuracy of 91.50% from Voting Ensemble model.

    View on Github

    Mobile Price Classification

    • The dataset contains mobile characteristics and price. The aim was to predict the price range of the mobile phone given its features. This was a multi-class problem.
    • I started by visualizing the data using pie charts, scatter plots and correlation heatmaps.
    • I then iterated models over the data in order to compare the accuracy. I ended up choosing an SVC model with a 91.75% accuracy.
    • Model explainability modules such as eli5 and pdp were used to try and understand why the model gave specific predictions, and what features had more impact on the price range of a mobile phone.

    View on Github

    COVID-19 Chatbot

    • I first created a dataset from common COVID-19 questions. The data file was in JSON format The dataset contained intents, patterns and responses.
    • I performed data cleaning techniques by removing stopwords, numbers, punctuation and html. I also performed lemmatization
    • I then created a bag of words that contained tokenized words. After which it was passed to a simple sequential model with Stochastic Gradient Descent as the learning rate.

    Contact Me

    FOR QUESTIONS AND PROPOSALS?

    REACH OUT TO ME

    Call Me

    +254721679735

    Location

    Nairobi

    Email

    nduatikevin1@gmail.com

    SEND ME AN EMAIL

    I'M VERY RESPONSIVE TO MESSAGES