Data science skills: Building partnership for efficient school curriculum delivery in Africa

1.Introduction

Data science is a “concept to unify statistics, data analysis, machine learning and their related methods” in order to “understand and analyze actual phenomena” with data. It employs techniques and theories drawn from many fields within the context of mathematics, statistics, computer science, and information science.

Data Science has spread its branches through several quintessential fields in modern day learning. It has emerged as a global phenomenon that has revolutionized industries and has increased their performances substantially [1]. Given the vast increase in the volume and complexity of data and the new technologies that have been developed to process and analyze this information, it can be argued that there is an increased need for statistical thinking in the context of working with data [2]. Key statistical reasoning topics that are critical for Data Scientists to know at a deep level include but are not limited to the following: developing clear statements of the problem/scientific research question; ensuring acquisition of high-quality data; understanding the process that produced the data, to provide proper context for analysis; allowing domain knowledge of the problem to guide both data collection and analysis; approaching modeling as a process that requires an overall strategy.

The modern day “romance” between Data Science and Statistics cannot be overemphasized (see Fig. 1). Statistics can be a powerful tool when performing the art of Data Science. From a high-level view, statistics is the use of mathematics to perform technical analysis of data. A basic visualization such as a bar chart might give some high-level information, but with statistics one gets to operate on the data in a much more information-driven and targeted way. The analysis involved helps to form concrete conclusions about our data rather than just guesstimating. Using statistics, we can gain deeper and more fine grained insights into how exactly our data is structured and based on that structure, optimally apply other data science techniques to get even more information [3].

Figure 1.

The interactive disciplines of data science.

Education is the key to shaping the lives of people. Since the dawn of civilization, humans have evolved through education and have developed mechanisms to improve education. In the 21st century, where data is omnipresent in every walk of life, education is no exception. With advancements in computing techniques, it is possible to imbibe all the information through powerful big-data platforms [4]. Various Schools have to keep themselves updated with the demands of the industry so as to provide appropriate courses to their students. Furthermore, it is a challenge for the Schools to keep up with the growth of industries. In order to accommodate this, Schools are using Data Science systems to analyze growing trends in the market [5]. Using various statistical measures and monitoring techniques, data science can be useful for analyzing the industrial patterns and help the course creators to imbibe useful topics. Furthermore, using predictive analytics, Schools can analyze demands for new skill sets and curate courses that address them [6].

The performance of students depends on the teachers. While there are many assessment techniques that have been used to assess the performance of teachers, it has been mostly manual in nature. With the breakthrough in data science, it is possible to keep track of the teacher performance. This is not only valid for recorded data but also real-time data. As a result, with real-time monitoring of teachers, rigorous data collection is possible, along with its analysis. Furthermore, we can store and manage unstructured data like student reviews on a big data platform.

2.Materials and methods

3.Results and discussion

The dataset was extracted in MS-excel and was saved as a “comma delimited (social.csv) file”. Another object was created in R for the social.csv file named w_africans as used in exporting the data into the console using the command line:

w_africans<-read.csv(“social.csv”,header=T)

However, the w_africans dataset was inspected for correctness before commencing the analysis using the commands stated below and the output is as given in Table 1.

#Displaying the first 15 observations of the w_africans dataset

print(head(w_africans, n=15))

The nature of the columns (variables) in the w_africans dataset was also explored, using

ls(DATAVAR) or names(DATAVAR), where DATAVAR represent the dataframe name to be explored using the commands given below, with the subsequent results.

#Dataset variable names can be viewed using names (dataset) or ls(dataset)

ls(w_africans)

[1] “Country” “ER” “GDPPC_PPP” “GNIPC_PPP”

#Viewing the number of rows and columns in the w_ africans dataset; use ncol(dataset) and nrow(dataset)

ncol(w_africans); nrow(w_africans)

[1] 5

[1] 320

From the results output, the w_africans dataset contains 4 variables and 320 rows as explained earlier

#A more advanced way to view the structure of the dataset is by using str(DATAVAR)

str(w_africans) #Data structure

data.frame’: 320 obs. of 5 variables:

$ Country: Factor w/16 levels “BF”,“BJ”,“CI”,..:2 2 2 2 2 2 2 2 2 2…

$ period: int 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008…

$ GDPPC_PPP: num 1622 1666 1703 1729 1735…

$ GNIPC_PPP: int 1260 1320 1380 1410 1450 1510 1540 1600 1690 1770…

$ ER: num 615 710 732 694 580…

The w_africans data.frame includes 2 numeric variables, 2 integer variables and 1 categorical variable

The Mean value of each of the variables is computed using the commands:

#Calculate the mean of variable with mean(DATAVAR$ VAR): mean of GDPPC_PPP variable

mean(w_africans$GDPPC_PPP, na.rm=TRUE)

[1] 2258.119

#mean of GNIPC_PPP variable

mean(w_africans$GNIPC na.rm=TRUE)

[1] 2117.962

#mean of ER variable

mean(w_africans$ER, na.rm=TRUE)

[1] 857.6926

Here, the average GDP at purchasing power parity per capita, GNI at purchasing power parity per capita and exchange rate (ER) for the 16 West African countries between years 1999 and 2018 is about $2258.12, $2117.962 and 857.6926 per US$ respectively.

Note: The na.rm =TRUE command the console to remove missing value in case there is one.

For the standard deviation, the following commands subsist; and the results represent the spread of the variables.

sd(w_africans$GDPPC_PPP, na.rm=TRUE)#Standard deviation of GDPPC_PPP

[1] 1331.402

> sd(w_africans$GNIPC_PPP, na.rm=TRUE)#Standard deviation of GNIPC_PPP

[1] 1341.855

sd(w_africans$ER, na.rm=TRUE)#Standard deviation of ER

[1] 1596.375

Continuing in the same terrain for the Range computation, minimum and maximum are computed on a single variable using the min(VAR) and max(VAR) formula. Students were taught how to calculate minimums and maximums using the codes below:

#Minimum and maximum GDP of the selected w_ african countries

min(w_africans$GDP, na.rm=TRUE); max(w_africans $GDP, na.rm=TRUE)

[1] 754.86

[1] 6661.99

From the output, the minimum GDP at purchasing power parity per capita is $754.86 and the maximum is about $6,661.99. This indicated a large gap in GDP per capita taking distribution among the West African countries in response to their purchasing power parity into consideration.

> #Minimum and maximum GNIPC of the selected w_ african countries

> min(w_africans$GNIPC_PPP, na.rm=TRUE); max (w_africans$GNIPC_PPP, na.rm=TRUE)

[1] 600

[1] 7330

It can be inferred that the Gross National Income at PPP per capital of all West Africa is between $600 and $7330 inclusive.

#Minimum and maximum ER of the selected w_african countries

min(w_africans$ER, na.rm=TRUE); max(w_africans$ ER, na.rm=TRUE)

[1] 0.27

[1] 9088.32

It is evidenced within the studied periods that Ghana’s economy has not been adversely affected by external forces as shown from their cedis minimum exchange rate to the US$ while the maximum exchange rate of 9088.32 is attributed to Guinea. We can infer that African countries as a nation is still developing and may take some time to meet up with other continents currency rates.

The command “range(VAR)” is used to summarize the minimums and maximums on individual variables. These computations are demonstrated in the following codes:

#Calculate the range of a variable with range(VAR)

range(w_africans$GDPPC_PPP, na.rm=TRUE)#Range of variable GDP

range(w_africans$GDPPC_PPP, na.rm=TRUE)#Range of variable GNDPPC_PPP

[1] 754.86 6661.99

range(w_africans$GNIPC_PPP, na.rm=TRUE)#Range of variable GNIPC_PPP

[1] 600 7330

range(w_africans$ER, na.rm=TRUE)#Range of variable ER

[1] 0.27 9088.32

Students have been taught that a quartile is a value computed from a collection of numeric measurements, showing observation’s rank when compared to all other present observations. Quartile can also be alternatively expressed as a percentilepercentile, as it is identical but on a scale of 0 to 100. Thus, we used quantile() function to obtain quartile and percentile in R, with commands

quantile(VAR, prob=c(prob value1, prob value2, …, prob valuei))

#Calculate the 25th, 50th, 75th percentilepercentile for GDP per capita at PPP

quantile(w_africans$GDPPC_PPP, na.rm=TRUE, prob =c(0.25, 0.50, 0.75, 0.95))

25% 50% 75% 95%

1369.780 1728.700 2851.580 5361.187

From the output, it easily observed that 25% of average GDP at PPP per capita was $136.780 with median (50th percentilepercentile) of $1728.700; 75% was about $2851.580 and 95% of the African countries have about $5361.187. This may explain the wide gap in GDP growth of each West African countries since GDP per capita is correlated with GNI per capita.

#Calculate the 25th, 50th, 75th percentilepercentile for GNIPC_PPP

quantile(w_africans$GNIPC, na.rm=TRUE, prob=c (0.25, 0.50, 0.75, 0.95))

25% 50% 75% 95%

1195 1680 2625 5435

#Calculate the 25th, 50th, 75th percentilepercentile for ER

quantile(w_africans$ER, na.rm=TRUE, prob=c(0.25, 0.50, 0.75, 0.95))

25% 50% 75% 95%

83.060 494.040 591.740 4528.037

Students were also taught how to use summary(x) function, where x can be any number of objects, including datasets, variables, and linear models to generate the descriptive statistics of the variables in the dataset. The code is written below for the w_africans dataset with the subsequent results presented below it.

> #Summarize the w_africans dataset using the command summary(x)

> print(summary(w_africans))

The summary outputs provides the descriptive statistics of all objects in the sample dataset and is explicitly presented in Table 2. Further exploration was carried out on the data by checking their respective distributions through Skewness, kurtosis and further test such as the Shapiro wilk test of normality. These were done using the “moments” library in R. Students were taught how to load packages from R as library(). Details are as given below while the summary presented in Table 3:

library(moments)

skewness(w_africans$GDPPC_PPP, na.rm=T) #Skewness coefficient of GDP per capita at PPP

[1] 1.353004

skewness(w_africans$GNIPC_PPP, na.rm=T) #Skewness coefficient of GNIPC at PPP

[1] 1.517567

skewness(w_africans$ER, na.rm=T) #Skewness coefficient of ER

[1] 3.283139

kurtosis(w_africans$GDPPC_PPP, na.rm=T) #Kurtosis coefficient of GDP per capita at PPP

[1] 4.226773

kurtosis(w_africans$GNIPC_PPP, na.rm=T) #Kurtosis coefficient of GNIPC at PPP

[1] 4.940481

kurtosis(w_africans$ER, na.rm=T) #Kurtosis coefficient of ER

[1] 13.80796

shapiro.test(w_africans$GDP)#GDP test of Normality

Shapiro-Wilk normality test

data: w_africans$GDPPC_PPP

W = 0.84758, p-value < 2.2e-16

shapiro.test(w_africans$GNIPC)#GNIPC test of Normality

Shapiro-Wilk normality test

data: w_africans$GNIPC_PPP

W = 0.83966, p-value < 2.2e-16

shapiro.test(w_africans$ER)#ER test of Normality

Shapiro-Wilk normality test

data: w_africans$ER

W = 0.5022, p-value < 2.2e-16

Positive coefficients of 1.353, 1.518, and 3.283 indicated that the econometric variables of GDP, GNIPC and ER is highly skewed to the right and may not be normally distributed. As the Kurtosis measure the fourth moments, selected West Africans exchange rate was found to be normally distributed (kurtosis ≈ 3) with other kurtosis of other variables > 3, indicating a leptokurtic shape compared to a normal distribution. However, normality test of the data confirmed the non-normality of the data since its associated p-values are lower than 5% level of significance.

Figure 2.

Normal Q-Q plots of GDP at PPP per capita of some selected West African countries.

Figure 3.

Normal Q-Q plots of GNI at PPP per capita of some selected West African countries.

Figure 4.

Normal Q-Q plots of ER of some selected West African countries.

Figure 5.

Bar chart of average GDP per capita based on PPP rates of selected West African countries.

Figure 6.

Bar chart of average GNIPC based on PPP rates of selected West African countries.

Figure 7.

Bar chart of average ER of selected West African countries.

Quantile plots visualize the distribution of the data per variable and details generated by the below commands are as given in Figs 2–4 respectively

par(mfrow=c(2,2)) #Partitioning of plots space

#Quantile plot of GDP per capita at PPP rates

qqnorm(w_africans$GDPPC_PPP);qqline(w_africans$ GDPPC_PPP,col=“red”)

#Quantile plot of GNI per capita at PPP rates

qqnorm(w_africans$GNIPC_PPP);qqline(w_africans$ GNIPC_PPP,col=“black”)

#Quantile plot of Exchange rate

qqnorm(w_africans$ER);qqline(w_africans$ER,col= “green”)#Quantile plot of ER

The Figs 2–4 showed that the quantile plots of the selected variables do not lie on the theoretical normal line. Thus, the variables are not precisely normal but may not be too far off.

Students were also introduced to data splitting in R using dataframe_name[n:m,]. This method was used due to the fact that the data structure was paneled in nature with the first 20 observations on row-wise which represents republic of Benin followed by Burkina Faso, among others. The command line used is given below with the results output presented in Table 4.

benin_d<-w_africans[1:20,];benin_d #Extracted Benin republic variables from the panel structured data.

The data was further explored using ExPanDaR package in R. Average GDP, GNIPC and ER per cross sections (countries) were visualized from the Shiny app using simple bar chart presented in Figs 3–5 respectively.

library(ExPanDaR)

ExPanD(df=w_africans)

The Figs 5–7 showed that Cape Verde (CV) recorded the highest average GDP (per capita) and GNI (per capita) taking into consideration purchasing power parity among the West African countries followed by Nigeria (NG). Cape Verde (CV) also has the highest average GNIPC at purchasing power parity rates and Ghana (GH) possess the strongest currency rate among other west African nations taking the US$ exchange rate into consideration. Niger (NE) recorded the lowest average GDP per capita and GNIPC at PPP and Guinea (GN) with the weakest currency rate within the selected timeframe. This can also be evidenced from Table 4 with an associated variability from the mean.

4.Conclusion

Introducing beginner students in statistics to data science is a vexatious task, especially in African countries where regular supply of power is a luxury and uninterrupted internet facilities are quite expensive and almost impossible. The developing nature of most Africa countries has created a paradoxical approach to achieving reasonable success in students’ learning of data science. However, for the purpose of this research, great achievement was made in introducing the students to data description using R software for data science, thereby equipping them with a career in data analysis. From the beginning, students offering introductory statistics gain reasonable experience of what constitutes both the practical and conceptual aspects of the working life of a data scientist, as they were able to run simple codes on exploratory data analysis using the focused data. The students equally enhanced their knowledge in deducing reasonable inference from the output of data analysis. 200 level students were able to run with ease, R codes to estimate basic descriptive statistics within a 1 hour lecture period. The activities was carried out without much supervision on the part of the tutor. Comparison was made per member countries on their developmental rate taking their respective Gross Domestic Product, Gross National Income per capita, and Exchange Rate into consideration.

It is of the opinion that topics covered in data science courses can and should be brought into a variety of statistics courses at undergraduate level, while adequate facilities provided for its teaching and learning. Thus, key data science skills need to be introduced, reiterated, and reinforced throughout the undergraduate statistics curriculum.

Though, the exercise is not without its own challenges, but its prospects in creating self-driven learning culture among students of tertiary institutions has greatly enhance the quality of teaching, advancing students skills in machine learning, improved understanding of the role of data in global perspective and on the spot ability of the students to be able to critique claims based on data.