Authors: Abduba Galgalo, Faith Makokha, Femi Kamau, John Mungai, Susan Mungai, and Willy Angole
This project is a part of the Data Science (DSF-FT) Course at Moringa School. The full project description can be found here.
Water makes up about 70% of the earth’s surface and is among the few basic needs common to all living creatures. Unfortunately, 96.5 percent of its coverage makes up oceans, which makes up for some scarcity on land. Tanzania is one of those countries that have to grapple with the scarcity of this life saving need. As a developing nation with a population of over 57,000,000, it has a hard time meeting the demand for safe drinking water due to limited resources for water extraction. There are already some water pumps available in the country but unfortunately, not all are functional and others need repair.
The World Health Organization- (WHO), understands that three out of ten people still lack access to clean water, highlighting the critical need for investment to improve water hygiene and accessibility. Water poverty has been considered to drive serious illnesses, high infant mortality rates, poor education, slumped economies and unproductive agricultural conditions in a majority of the regions.
Tanzania is currently failing to meet the potable water requirements for its population. Despite already building water wells to offset this problem, many are unfortunately non-functional and others are in need of repair. As a result, our stakeholder, WHO, has narrowed down their interest to drilling more water wells and maintaining/ revamping existing ones within the country, in an effort to ensure availability of quality and quantity drinking water countrywide .
Our role as the data scientist in this project will be to identify patterns in non-functional wells, with the aim of influencing how new ones are built. Furthermore, using these patterns, we will enable our stakeholder to accurately predict existing water points in need of intervention ensuring the people of Tanzania have access to clean potable water
- To identify the trends/patterns between both non-functional and functional wells
- To identify non-functioning wells using a simple analysis, and predict the functionality of a well based on available variables.
- How can we apply Machine Learning and necessary classification methods to predict functionality of wells in Tanzania?
- Are there any common features that are consistent with functional and non-functional wells?
- Can we identify ways to help build better wells?
- To ensure that newly constructed wells are of good quality water for the communities.
- To correctly identify functionality of a well and determine its viability.
- To successfully predict the type of wells that dry up and determine the type of wells to build based on the area?
- To determine working wells in different regions and which ones need repair or improved management.
- To ensure built wells are longer lasting/ more viable.
Generating a model that will be able to correctly predict the quality status of the wells in Tanzania with an accuracy of 80%.
In this project we shall use a dataset containing information about existing water wells in Tanzania sourced from an ongoing DrivenData competition.
The dataset contains 59,400 records and spans 40 columns. Of these columns, we identified 31 to be categorical, and 9 as numerical. We were able to further group the columns into the general features being captured. Some of the rows include:
Numerical
- amount_tsh - Total static head (amount water available to waterpoint)
- longitude - GPS coordinate
- latitude - GPS coordinate
- gps_height - Altitude of the well
- population - Population around the well (those living near it)
Categorical
- extraction_type - The kind of extraction the waterpoint uses
- extraction_type_class - The kind of extraction the waterpoint uses
- extraction_type_group - The kind of extraction the waterpoint uses
- water_quality - The quality of the water
- quality_group - The quality of the water
- quantity - The quantity of water
- quantity_group - The quantity of water
- source - The source of the water
- source_type - The source of the water
- source_class - The source of the water
- waterpoint_type - The kind of waterpoint
- waterpoint_type_group - The kind of waterpoint
- status_group - The condition of the wells (target variable)
his dataset’s analysis and classification will help our stakeholders to improve maintenance of the present water wells or give useful information for future wells.
There are a lot of repetitive columns in the dataset, for example. payment and payment_type, source and source_type which have the same information. We scrapped the columns and for the sake of preparing our data for modelling purposes.
We have included columns that contain information about the location and the basin that is closest to the water wells geographically.
The data was checked for missing values and a number were found in the columns; funder, installer , subvillage, scheme_management and permit the percentage of missing values in the particular columns were a small amount, thus were dropped to ensure completeness of the dataset.
Checking for validity in the data, the dataset was checked for any duplicated values and outliers. The duplicated records were not dropped as it does not mean that they were similar wells but just built under the same project.
We also chose not to drop the outliers, as it did not display erroneous data but will be further looked into in the analysis section.
After cleaning the dataset, we need to bring uniformity by formatting and the columns to be readable and easily interpretable.So we defined functions to make these possible.
Moreover, we verified that the values of various columns are consistent. The names of installers and funders had numerous instances of the same funder and/or installer, but with misspelt names and/or typographical errors.
To begin, we point out that our dataset is massive and that there is class imbalance; specifically, 54.1% of our data is considered to be "functional," while 38.8% is "non-functional," and the remaining 7% pertains to "non-functional wells that need repair." We One-Hot-Encoded the categorical variables and categorised the target. Since the majority of our information comes from operational wells, maintaining the current imbalance isn't a priority for us. Our baseline model was LogisticRegression with the StandardScaler as our scaler, just to test how our chosen attributes would perform on a basic level. We then explored other more sophisticated models like the RandomForest, Decision trees, K-Nearest -Neighbours and XGBoost.
We used a pipeline to scale the data and then fit it to the model. We then used cross validation to ensure that our model was not overfitting. We also used a confusion matrix to evaluate the performance of our model.
Despite the fact that our models did not achieve the desired accuracy of 75%, we were able to achieve an accuracy of 70% which is a good start for our project and is within an acceptable range of +/- 5%
The accuracy of both our random forest classifier and decision tree models was 70%. While this is still a good predictive model, we would like to undertake further feature engineering to boost this recall score if we had more time. We achieved our objectives to be able to predict the functional wells and had an acceptable accuracy score.
When our stakeholder decides to construct additional wells in Tanzania, they may want to look into the Lake Rukwa as a basin area, where there are disproportionately more non-functional wells than functional ones. The region of Dodoma has more non-functional wells than functional, this area needs to be looked into. Wells permitted to operate tend to be more viable and functional over time than those without. The wells that are not paid for tend to be non-functional as they are maybe misused by the public, maybe implementing an affordable payment scheme will help curb this. Wells without permits also have a higher chance to be non-functional so this means that our stakeholder needs to make sure that they are permitted to ensure they are suitable for human consumption as well. Wells with close proximity to Lake Victoria basin tend to be long-lasting compared to the rest of the basins.
- The repository is structured as follows:
├── README.md <- The top-level README for reviewers of this project
|
├── .gitignore <- The .gitignore file for the project
|
├── LICENSE.md <- The license for the project
|
├── index.ipynb <- Narrative documentation of analysis in Jupyter Notebook
|
├── modeling.ipynb <- Narrative documentation of modeling in Jupyter Notebook
|
├── cleaned_data.csv <- The cleaned dataset
|
├── presentation.pdf <- PDF version of project presentation
|
├── write_up.pdf <- PDF version of project write up
|
└── data <- Datasets used in the analysis
├── well_data_labels.csv
|
└── well_data_values.csv
- Python
- Pandas
- Matplotlib
- Sklearn
- Please review the full analysis in the Jupyter Notebook or the Presentation.