Hi, I'm Abigail

Abigail Jatto

Data Scientist

Data Scientist with a background in biochemistry, pharmaceutical quality assurance and customer analytics. I am currently completing an MSc in Artificial Intelligence and Data Science at the University of Hull where I build end-to-end machine learning pipelines, deep learning models and data visualisations across real-world datasets spanning road safety, healthcare, NLP and customer behaviour.

Presently, I am optimizing transformer-based models such as Roberta to explore consumer sentiment toward AI-generated advertising and its impact on brand perception using NLP techniques to analyse real-world discourse across multiple marketing campaigns.

Previous projects led to measurable outcomes including a 30% improvement in business growth, a 96.7% accurate churn classifier using Power BI and multi-domain recommendations drawn from 370,000 UK road accident records. I bring a rigorous, data driven approach shaped by experience in regulated pharmaceutical environments, academic research and fast paced technology organisations.

View my work ↗

Projects

Selected work

Analyses and dashboards across road safety, customer behaviour, and business performance.

Featured project
370K
accident records
7
analyses
3
years of data

Python · Data Science

UK Road Traffic Accident Analysis (2017 to 2019)

Analysed 370,153 DfT STATS19 accident records to uncover the conditions and patterns driving UK road accidents, using temporal analysis, geospatial clustering, association rule mining, and time series forecasting.

DBSCAN SARIMA Apriori scikit-learn statsmodels pandas
  • Serious accidents associate with fine weather & dry roads, not bad conditions
  • Peak risk: 08:00 and 15:00 to 18:00 weekdays (commuting, not weather)
  • Bradford & Leeds identified as top West Yorkshire accident hotspots
View on GitHub ↗ Notebook ↗
Featured project
120K
news articles
4
models compared
32
configurations tested

Python · Data Science · NLP

Automated News Text Classification — AG News Dataset

Compared 4 classification approaches from Naïve Bayes to Bidirectional LSTM on 120,000 news articles across 4 categories, executing 32 hyperparameter configurations to identify where and why performance gains occur across the full ML to deep learning spectrum.

Naïve Bayes LinearSVC Word2Vec CNN BiLSTM TF-IDF scikit-learn TensorFlow
  • Accuracy progression: 88% (Naïve Bayes) → 89% (SVM) → 91% (CNN) → 92% F1 (BiLSTM)
  • Deep learning advantage over well-tuned classical models: real but modest at 3 to 4 points
  • GridSearchCV across 24 combinations for classical models; randomised search across 8 for neural architectures

Excel · Data Visualization

Coffee Sales Dashboard

Analysed sales data across multiple coffee stores using Excel, identifying highest performing stores, top products, and seasonal trends.

ExcelDashboardSales Analysis
Read article

Power BI · Customer Analytics

Databel Customer Churn Analysis

Analysed Databel customer data using Power BI to identify churn drivers and surface growth opportunities with actionable recommendations.

Power BIChurnSegmentation
View dashboard

SQL · Data Cleaning

Film Dataset Cleaning

Duplicated, cleaned, and prepared a films dataset for downstream analysis, handling missing values, inconsistent formats, and data quality issues in SQL.

SQLData CleaningETL
View on GitHub

Tools & Languages

Python SQL Excel Power BI Jupyter Notebook pandas numpy matplotlib seaborn scikit-learn statsmodels

Analysis Skills

Time series forecasting Clustering Association rules EDA Data visualisation Customer segmentation Churn analysis Data cleaning

Contact

Get in touch

I'm open to data science roles, collaborations, and interesting projects. Feel free to reach out on any of the platforms below.

LinkedIn abigail-jatto GitHub Abigail-Jatto Medium @jattoabigail548