Highlights

AMHARIC TO ETHIOPIA SIGN LANGUAGE TRANSLATION

Method Used:

• Collect and preprocess training data.

• Build an application that displays the translated sentences in graphics animation.

• Deploy Grammer translation and Speech recognition models.

Machine Learning:

• A pre-trained model for word Lemmatization.

• A speech recognition system.

• A grammar-translation system, from Amharic to Ethiopian sign language

Technologies used:

• Python, C#, TensorFlow, Keras, Flask, UnityGameEngine and Blender.

TWITTER COVID DATA ANALYSIS

Preprocessing:

• Analysis of covid data scraped from Twitter following the CRISP-DM methodology.

Machine Learning:

• Topic modeling

• Sentiment Analysis

Technologies used:

• Python, MySQL, NLP libraries, StreamLit, Docker and Heroku

AMHARIC SPEECH RECOGNITION

Preprocessing:

• Audio data preprocessing like resampling, normalization, and removing noise.

• Perform data augmentation by adding noise, changing the speed and pitch of the audio.

• Implemented audio data feature extraction method using log Mel spectrogram.

Machine Learning:

• Build a model with two main neural network modules. Three layers of Residual Convolutional Neural Networks to learn the relevant audio features. Then a set of Bidirectional Recurrent Neural Networks to leverage the learned audio features.

Technologies used:

• Python, TensorFlow, Keras, NLP libraries, DVC, MLflow, and StreamLit

COVID-19 IMPACT ON DIGITAL LEARNING

EDA:

• Applied Data wrangling techniques on telecommunication data.

• Performed exploratory data analysis.

• Conducted univariate and multivariate graphical analyses using Seaborn and Plotly

Machine Learning:

• Applied time series based clustering to find out which communities responded similarly using Tslearn library.

• Analyzed the impact of school closure on states with the most and least pct_black/Hispanic; to check if covid 19 has disproportionately impacted student engagement with online learning platforms in areas where there are more Black or Hispanic Students using CausalImpact library.

Technologies used:

• Python, Pandas, Seaborn, Plotly, Tslearn, and CausalImpact

CAUSAL INFERENCE

EDA:

• Applied Data wrangling techniques in Breast Cancer Wisconsin (Diagnostic) data available on Kaggle.

• Exploratory analysis was carried out to observe features that have a higher correlation with the diagnosis.

• Performed feature extraction and scaling

Machine Learning:

• Perform feature descritization using a supervised learning approach.

• Used CausalNex to develop Bayesian Network models, that go beyond correlation and consider causal relationships.

• Trained a Logistic Regression model with the entire dataset and features.

Technologies used:

• Python, CausalNex, and Scikitlearn

SALES PREDICTION

EDA:

• Data exploration using Pandas, Matplotlib, Numpy. Modular code.

• Creation of new features.

Machine Learning:

• Applied Random Forest model that takes multiple variables as an input and predicts sales

• Applied an LSTM recurrent neural network that takes six weeks of historical sales data and makes predictions for future sales.

• Calculated Feature Importance to see which variables are mainly responsible for affecting the number of Sales and Customers.

• Deployed TensorFlow model in a production environment with StreamLit dashboard.

Technologies used:

• python, Scikitlearn, Tensorflow, DVC, MLflow, StreamLit, and Docker

A/B TESTING

Metric Choice:

• Invariant metrics: Used this to ensure that the experiment (the way we presented a change to a part of the population) is not inherently wrong. E.g. Number of users in both groups.

• Evaluation metrics: metrics we expect to change and are relevant to the goals we aim to achieve egg (brand awareness) Hypothesis testing for A/B testing.

• We use hypothesis testing to test the two hypotheses:

  • Null Hypothesis: There is no difference in brand awareness between the exposed and control groups in the current case.

  • Alternative Hypothesis: There is a difference in brand awareness between the exposed and control groups in the current case.

Machine Learning:

• Carried out 3 types of classification analysis to predict whether a user responds yes to brand awareness, namely: Logistic Regression Decision Trees XGboost, then compared the different classification models to assess the best performing one(s).

Technologies used:

• Python, Scikitlearn, XGBoost, SciPy, DVC, MLflow, StreamLit, Docker, and Heroku

PYTHONLIDAR

Features:

• Download point cloud data from the EPT resource on AWS cloud storage.

• Terrain visualization

• Data transformation

Technologies used:

• Python, PDAL, Laspy, Geopandas, Pydocs, and Heroku

TELECOM DATA ANALYSIS

EDA:

• Applied Data wrangling techniques on telecommunication data.

• Exploratory analysis was carried out to observe customer behavior in the telecommunication industry.

• Perform dimensionality reduction using PCA.

• Performed self-explanatory visualizations using tools such as plotly, seaborn, and matplotlib to get rich insights to improve customer experience and reduce the churn rate.

• Metrics like Experience Analysis, Satisfaction Analysis, Engagement Analysis were computed.

• Provided comprehensive report on the analysis to management for decision making.

Machine Learning:

• Perform clustering of the customers using the k-means clustering algorithm.

Technologies used:

• Python, Scikitlearn, SciPy, StreamLit, Docker, and Heroku