Which Water Pumps are Faulty?

Water Pump Maintenance Prediction in TanzaniaOverviewThere is a water crisis in Tanzania. Natural freshwater sources in Tanzania such as lakes, rivers, streams, dams, and groundwater are not evenly well distributed all over the country. According to water.org, only 50 percent of Tanzanias population of 53 million have access to an improved source of safe water.

The rest has to rely on either ground or surface water extraction. However, as the drainage system is poorly constructed, the leakage into the groundwater system becomes a major source for contamination; only 34 percent of Tanzanias population has access to improved sanitation. Besides that, almost half the water supply is wasted through leaks.

Under these circumstances, people, particularly women and young girls, spend a significant amount of time traveling some distance to collect water. This prevents young girls from attending schools. Therefore, a water crisis is also aggravated by the long term decrease in sustainability in future generations in Tanzania.

Problem StatementIn Tanzania, nearly half of installed water pumps since 1970, are non-functional or need repair. According to many studies, the primary reasons for this failure are:- Poor quality of material used in water pumps construction;- Inaccurate placement of groundwater extraction boreholes.- The shortage of skilled labor.This project aims to point out potential causes for water pumps malfunctioning, and in this way to predict the maintenance requirements of the water pumps, and to optimize resources allocation to increase populations access to freshwater supply and improved sanitation. On the plot above, each colored dot represents a water pump and its condition.

We can see that water pumps are clustered in specific areas of the country; there is a higher proportion of red dots, the non-functional water pumps, in the southeast and green dots, the functional water pumps, in the central and southwest parts of the country.The datasetIn this project, we will be looking into the dataset obtained through the DrivenData and attributed to the Taarifa waterpoints dashboard, which aggregates data from the Tanzanian Ministry of Water. The initial dataset contains 59400 observations and 40 features. Out of 40 features, 30 are categorical, 8 are numerical, and 2 are date features.

The features contain information on water source types, payment types, funder, installer, geographical location, construction year, date records taken, etc. The target category this project aims to predict is the condition of water pumps, which are either of:-functional,-non-functional,-functional, but needs repair.Now, lets check the count of each condition of the pumps in the target category so that we have a big picture of the overall situation:From the graph above we see that nearly half of all water pumps are either non-functional or functional needs repair.Lets check the number of projects constructed by year:The graph above shows growth for the number of water pumps installed over the years. As we look back in time, we see the proportion of non-functional water pumps increases.

The reason is most probably poor maintenance.In different regions, the proportion of functional / non-functional water pumps varies:Water pumps condition by payment type:Well, too sad, when there is no payment for a water pump, its not been taken care of, which clearly shows the proportion of non-functional water pumps to functional ones. Model Selection and EvaluationThe initial dataset included many duplicate features and was incomplete in a way that many important features were filled with null values. After removing duplicates and features with the high number of null values, and adding 3 synthetical features, 22 features remained in the resulting dataset.

The question the model should answer is, which class a specific water pump would belong to, i. e is a random water pump constructed in X year, funded by Y organization, located in Z area, and etc. would be functional, non-functional, or functional needs repair?

An effective model developed in this project aims to optimize resources allocation to increase population access to freshwater and improved sanitation.In this project, accuracy was used as a measure of prediction. Accuracy is defined as the fraction of correct predictions compared to the total predictions.Majority Class BaselineFast and first baseline model to understand the minimum that any following machine learning model shouldnt hit. For this approach, the majority class was a mode of the condition of water pumps, status_group label in the dataset.

For this model, the test data accuracy score is:MCB Test Accuracy Score: 0.5430Logistic RegressionLogistic regression outputs a probability that the given input point belongs to a certain class. Its named for a function used at the core of the method, the logistic function. Here, the test data accuracy score is:LR Test Accuracy Score: 0.7450Random Forest ClassifierRandom Forest Classifier is an ensemble algorithm.

Random forest classifier creates a set of decision trees from a randomly selected subset of a training set. It then aggregates the votes from different decision trees to decide the final class of the test object. The test data accuracy score received:RFC Test Accuracy Score: 0.8114So far, the Random Forest Classifier yielded the highest accuracy score on the test data.

ConclusionThe conclusion is that there is a potential for better use of funds allocated for water pumps projects in Tanzania. What the projects need, in addition to funds, are skilled professionals to investigate materials appropriate for the climate and water quality to build more sustainable water pumps; better maintenance techniques and procedures; and in connection with that, help people in local areas to get a better education on how to build/maintain water pumps RELATED QUESTION Why do American men wear class rings, but wont use their postnominal degrees on door signs and business cards? American women wear class rings, too.

I wear mine.Generally speaking, displaying where you got your degree from on a business card or a door sign would be seen as odd, at best. If somebody handed me a business card with the Harvard logo on it and he didnt work at Harvard, Id probably ask about it: Oh, do you do work with Harvard University? and if the answer was, No, I just attended Id find this an unpleasant level of showing off.

Not classy.A lot of Americans will have college-related paraphernalia in their offices, though. Like, you may see a pennant or they may have their degree(s) up on the wall. This is a more casual display of affiliation.Likewise, the ring is casual.

Class rings are usually not that obtrusive, particularly the signet variety. Usually, the signet is only immediately recognized by others who attended the university (though I have had people ask me to hand over my ring so they could take a closer look at it before), and thus it kind of acts like a secret handshake. Ive had people from my college recognize my signet ring on sight, and then we have an instant connection.Its just a lot classier than shoving it in everybodys face all the time in an American social environment, basically.