Hello Broadcast,
Welcome back to DSAIL Mail! This edition is full of exciting updates. We co-organized this year's Data Science Africa, celebrated several members getting their workshop papers accepted and even had a team member win a Kaggle competition. Though we bid farewell to one of our own who relocated to Switzerland, we have gained some new additions to the team. Buckle up!
|
Data Science Africa (DSA)
Teamwork |
At DSAIL, collaboration is at the heart of our mission. Recently, we played a pivotal role in co-organizing Data Science Africa (DSA), which brought together minds passionate about leveraging data for societal impact. We extend our gratitude to everyone who joined us at this event, which promises to inspire future innovations in our field. |
|
|
Data Science Africa (DSA)
Grevy’s Zebra and Plains Zebra Species Identification: A Case Study of Mugie Conservancy in Northern Kenya |
Workshop Paper
Camera trapping has been widely adopted as a way of monitoring wildlife in their natural habitat due to their non-invasive nature. Camera traps deployed strategically in high-concentration areas such as wildlife corridors can generate enormous amounts of data. This makes manual analysis of camera trap datasets infeasible. Our paper focuses on the use of computer vision techniques to analyse camera trap image datasets. Specifically, we aim to detect and distinguish between two species of zebras: the Grevy's zebra, which is an endangered species endemic to Northern Kenya, and the common Plains Zebra.
We trained YOLOv8 object detection models using data collected by conservationists at Mugie Conservancy in Laikipia County. The models were evaluated against a manually annotated dataset and achieved a human-level accuracy in annotating 68.4% of the images. The results show promise in improving the efficiency and accuracy of identifying different species, which is crucial for conservation purposes. |
Data Science Africa (DSA)
Prediction of Coffee Yield with Time Series Modeling and Ensemble Techniques |
Workshop Paper
This paper, by Teofilo Ligawa and Cedric Kiplimo, dives into the accurate forecasting of coffee yields using environmental and farm-specific data. Our research focuses on delivering actionable insights crucial for strategic agricultural planning and global food security.
We have leveraged a mix of models like random forest, XGBoost, Facebook Prophet, and LSTM networks to analyse data patterns and relationships. Through a combination of ensemble and time series forecasting techniques, we achieved impressive results. Our Multistep 1 step ahead model stood out with an RMSE of 181.60 kg, showcasing its robustness in predicting coffee yield outcomes. |
|
|
Data Science Africa (DSA)
Precipitation Quality Control Methods For A Large Weather Network in Africa |
 |
Workshop Paper
Addressing the challenge of sustaining food production amidst depleting groundwater and a growing population, reliable weather data is essential, especially in Africa, home to 65% of the world’s remaining uncultivated arable land. The Trans-Africa Hydro-Meteorological Observatory (TAHMO) has established an IoT network to collect vital environmental data for optimising crop yields. However, ensuring data quality, particularly for precipitation, remains a challenge.
Our team member, Austin Kaburia, presented a data pipeline design that includes the RainQC regression model, improved with a median method for quality control. These methods significantly enhance the detection of faulty rain gauges. The paper also addresses scalability and future integrations, including Gaussian Process Spatial Anomaly Detection and weather forecasting, to further improve data quality.
This paper and the preceding papers were accepted by Data Science Africa (DSA) and will be available on our website with our other publications soon. |
|
|
Data Science Africa (DSA)
PhysioNet Challenge |
Data Science Africa (DSA)
Bioacoustic Fieldwork |
Fieldwork
During DSA, DSAIL members presented a session on running ML on edge devices together with Sarah, Ross and Mathijs a team from Edge Impulse, focusing on deploying ML for ecosystem monitoring. Fieldwork involved deploying acoustic sensors at DeKUT Wildlife Conservancy to classify birds by vocalisation using loaded ML models. There was a lecture session introducing concepts and another session analysing and discussing results post-field exercise. |
|
|
Competitions
BirdCLEF 2024 Competition |
DSAIL research intern, Yuri Njathi, participated in Kaggle’s BirdCLEF 2024 competition winning a bronze medal for his participation and unlocking Kaggle’s competition ranking to find himself ranked 935 of 204,000 (among the top 0.5%) data scientists on Kaggle. |
|
|
Next Chapter
Bidding Farewell to Dishan |
Spotlight
Meet Our New Interns |
Listen to what our new interns have to say: |
|
|
Paul Bett
My first month at DSAIL has been incredibly rewarding. It is my first time in Nyeri, my first DSA attendance, and my first encounter with ML and AI enthusiasts from other countries. I am involved with TAHMO, performing quality checks to enhance African environmental sustainability. This journey is one I am thrilled to be a part of. |
|
Excited to join the DSAIL team at such an opportune time to attend my first DSA conference, which provided an opportunity to meet and interact with data science enthusiasts from different parts of the world. I am passionate about AI in healthcare and am currently working on a pose estimation AI solution for limb alignment in orthopaedic care. |
|
Inspired by life, I am developing autonomous agents to thrive in extreme and everyday environments. Attending DSA 2024 in Nyeri, where I learned about embodied AI and fine-tuning, fueled my excitement. The support from Prof. Ciira, Jason, Cedric, and the team makes this journey incredible.
|
|
|
Did a friend forward this to you? You can subscribe here
|
|
|
|
|
