| Course Code | ISYS3447 |
|---|---|
| Course Name | Introduction to Business Analytics |
| Title of Assignment | Assessment 3: Interactive Presentation and Reflective Journal |
| Student Name & Student Number | Bui Minh Tri – S4022413 |
| Teacher’s Name | Mr. Hoang Dinh |
| Assignment Due Date | 11:59 PM, September 2026 |
| Date of Submission | 18 September 2026 |
Absenteeism time in hours is a key indicator of employee absenteeism; frequent or prolonged absences can disrupt workflow, diminish morale, and reduce productivity and operational efficiency, undermining overall workforce stability (Nuruzzaman and Raihan 2026). Absenteeism is worth investigating within any organization as there are many factors, both internal and external can be associated with long absences - personal characteristics, work conditions, or the timing and reason for an absence. Therefore, identifying these factors allows organizations to move from reactive to proactive in addressing the problem. Understanding employees’ absenteeism patterns according to these personal and work metrics can help organizations build effective prediction models and design targeted solutions to sustain performance and ensure effective human resource planning (Liang and Li 2025).
A preliminary review identified several data-quality issues that could affect the analysis. These issues were addressed before conducting the descriptive and predictive analyses.
| Data issue | Treatment |
|---|---|
| 34 exact duplicate records | Removed to prevent double-counting. |
| Invalid categorical codes | Unsupported values in Day of the week, Month of absence, Seasons,
Education, Disciplinary failure, Social drinker and Social smoker were
treated as missing (NA). |
| Zero recorded absenteeism hours | Retained because there was insufficient evidence to classify these observations as errors. |
| Extreme absenteeism values | Reviewed as potential outliers and retained because they may represent legitimate extended absences. |
| Missing values | Retained as NA; available observations were used where
appropriate for each analysis. |
These treatments were applied to improve data consistency while avoiding unsupported assumptions about potentially valid employee records.
The distribution of absenteeism is strongly right-skewed, with the
majority concentrated at relatively low levels. Absenteeism ranges from
0 to 122 hours, with a median of 4 hours and a mean of 7.11 hours. This
indicates that while most employees record relatively limited absence, a
smaller number of observations experience substantially longer periods
of absenteeism.
Absenteeism is concentrated in several absence reasons. Reasons 13, 11, 19 and 12 accounted for the largest total number of absenteeism hours, with 1,453, 1,419, 1,414 and 1,350 hours respectively. These statistics indicate that absenteeism is not evenly distributed across reasons, providing a foundation for HR to examine the organisational and employee circumstances associated with the most common contributors.
Average absenteeism varies across months and days of the week,
demonstrating that absence patterns are not evenly distributed over
time. The heatmap highlights periods with relatively higher or lower
average absenteeism, which is useful for HR to identify recurring
temporal patterns for workforce planning. These patterns are interpreted
descriptively rather than as evidence that a particular month or weekday
causes absenteeism.
Absenteeism also varies across age–service profiles. Employees who
are aged beyond 50 with more than 15 years of service show particularly
high average absenteeism at 14.0 hours (n = 121), while those under 30
with 0–4 years of service also show relatively elevated absenteeism at
8.0 hours (n = 108). These differences highlight the relevance of
considering employee age and service-time profiles when examining
absenteeism patterns (White et al., 2015).
Across most of the top 10 absence reasons, the 30–39 age group represents the largest share of absenteeism hours. However, reason 12 is an exception, with employees beyond 50 accounting for the largest share. This indicates that the age composition of absenteeism differs across reasons while simultaneously suggesting HR interventions need to consider both the type of absence and the employee profile rather than applying the same approach.
The correlation analysis illustrates that no individual numeric predictor has a strong linear relationship with absenteeism. Age is the strongest association among all variables, however, it remains weak, with a positive correlation of r = 0.07, while distance has the largest absolute correlation at r = −0.09. This suggests that absenteeism is unlikely to be explained by any single numeric factor and supports the use of a multivariable approach to examine predictors simultaneously (Miraglia and Johns 2021).
The predictive analysis examines whether employee and workplace characteristics provide useful information for predicting absenteeism time. Four models were developed and compared using a consistent train-test approach: multiple linear regression, log-transformed linear regression, negative binomial regression and Random Forest.
From on the descriptive findings, four predictive models were developed to examine whether employee and workplace characteristics could help predict absenteeism hours. The models included Multiple Linear Regression, Log-Transformed Linear Regression, Negative Binomial Regression, and Random Forest Regression. The predictors were age, service time, workload, distance from residence to work, education, disciplinary failure, social drinking and social smoking. Previous research suggests that absenteeism can be influenced by a combination of personal, demographic, job and organisational factors. Therefore, using multiple predictors is appropriate to examine potential factors associated with absenteeism (Čikeš et al., 2018).
The models were compared using Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and Predictive R². MAE shows the average size of prediction errors, while RMSE gives more weight to larger errors. Predictive R² shows how much variation in absenteeism can be explained by the model when applied to the test data. These measures are particularly relevant to this dataset because absenteeism is highly right-skewed, ranging from 0 to 122 hours, with a mean of 7.11 hours and a median of 4 hours.
model_1 <- lm(
`Absenteeism time in hours` ~
Age +
`Service time` +
`Work load Average/day` +
`Distance from Residence to Work` +
Education +
`Disciplinary failure` +
`Social drinker` +
`Social smoker`,
data = model_data
)
model_2 <- lm(
log1p(`Absenteeism time in hours`) ~
Age +
`Service time` +
`Work load Average/day` +
`Distance from Residence to Work` +
Education +
`Disciplinary failure` +
`Social drinker` +
`Social smoker`,
data = model_data
)
library(MASS)
model_nb <- glm.nb(
`Absenteeism time in hours` ~
Age +
`Service time` +
`Work load Average/day` +
`Distance from Residence to Work` +
Education +
`Disciplinary failure` +
`Social drinker` +
`Social smoker`,
data = model_data
)
model4_train <- randomForest(
x = rf_x_train,
y = rf_y_train,
ntree = 500,
importance = TRUE
)
pred_model4 <- predict(
model4_train,
newdata = rf_x_test
)
| Model | MAE | RMSE | Predictive_R2 |
|---|---|---|---|
| Model 1: Multiple Linear Regression | 5.350 | 10.993 | 0.012 |
| Model 2: Log-Transformed Linear Regression | 4.567 | 11.294 | -0.043 |
| Model 3: Negative Binomial Regression | 5.293 | 10.952 | 0.019 |
| Model 4: Random Forest Regression | 4.998 | 10.532 | 0.093 |
The results show some differences between the four models. Random Forest Regression produced the lowest RMSE (10.532 hours) and the highest Predictive R² (0.093), while its MAE was 4.998 hours. This indicates that it performed better based on RMSE and Predictive R². However, the Log-Transformed Linear Regression had the lowest MAE at 4.567 hours, showing that it had the smallest average prediction error. Its Predictive R² was −0.043, however, suggesting that it provided limited predictive value when evaluated on unseen data.
Overall, the results suggest that the available employee and workplace variables provide some, but limited, information for predicting absenteeism. The Random Forest model performed best based on RMSE and Predictive R², but its Predictive R² of only 0.093 indicates that most of the variation in absenteeism remains unexplained. This is consistent with research showing that absenteeism is influenced by a wide range of factors, suggesting that other relevant variables may not be captured in this dataset because they were not available (Miraglia and Johns, 2021). Based on these findings, predictive modelling should be used as a supporting tool for workforce planning and identifying patterns for further review, rather than as the basis for decisions about individual employees.
Overall, absenteeism varies across absence reasons and employee profiles, while individual numeric variables, including workload, show only weak linear relationships with absenteeism. The predictive analysis also provides limited information from the available employee and workplace variables, suggesting that data analysis should be used as a supporting input for HR workforce planning rather than as a standalone decision-making tool.
1. Integrate absenteeism patterns into workforce planning
HR Operations should prepare a monthly absenteeism summary covering major absence reasons, employee-profile patterns and relevant workload patterns. The summary should be provided to the HR Manager and department managers to support reviews of staffing levels, workload allocation and recurring absence patterns. Where repeated patterns are identified, managers should review relevant work arrangements and staffing conditions rather than assuming that a single factor causes absenteeism (Anzoom et al. 2026; Miraglia et al. 2025).
2. Use predictive modelling as a decision-support tool
HR Analytics should use the selected predictive model to identify broader workforce patterns associated with higher predicted absenteeism and provide monthly insights to the HR Manager. Department managers can use these insights to review staffing, workload and work-design conditions where relevant. Given the model’s limited Predictive R², predictions should support HR judgement and workforce planning rather than be used as a basis for individual employee decisions (Lawrance et al. 2021; Notenbomer et al. (2019).
3. Strengthen absenteeism data governance
HR Operations should conduct a monthly data-quality review of absenteeism records, focusing on key categorical and employee-information fields before reports are produced. Data-quality exceptions should be documented and reported to the HR Manager for correction. This process would improve the reliability of future absenteeism analysis and provide a stronger data foundation for predictive modelling (Lawrance et al. 2021).
Anzoom R, Ali SM, Bhattacharjee P, Kabir G and Paul SK (2026) ‘Predicting employee absenteeism: A framework and its validation’, Array, 30, doi:10.1016/j.array.2026.100766.
Čikeš, V., Maškarin Ribarić, H. and Črnjar, K. (2018) ‘The determinants and outcomes of absence behavior: A systematic literature review’, Social Sciences, 7(8), p. 120. doi:10.3390/socsci7080120.
Lawrance A, Petrides G and Guerry M (2021) ‘Predicting employee absenteeism for cost effective interventions’, Decision Support Systems, 147, doi:10.1016/j.dss.2021.113539.
Liang X and Liang J (2025) ‘Sustainable human resource management and employee performance: a conceptual framework and research agenda’, Human Resource Management Review, 35(2), doi:10.1016/J.HRMR.2024.101060.
Miraglia M and Johns G (2021) ‘The Social and Relational Dynamics of Absenteeism From Work: A Multilevel Review and Integration’, Academy of Management Annals, 15:37–67, doi:10.5465/annals.2019.0036.
Notenbomer A, van Rhenen W, Groothoff JW and Roelen CAM (2019) ‘Predicting long-term sickness absence among employees with frequent sickness absence’, International Archives of Occupational and Environmental Health, 92: 501–511, doi:10.1007/s00420-018-1384-6.
Nuruzzaman M and Raihan T (2026) ‘Understanding and Combating Modern Slavery in the Ready-Made Garments Industry in Bangladesh’, Journal of Contemporary Asia, 56(2):284–302, doi:10.1080/00472336.2025.2485422.
White MI, Wagner SL, Schultz IZ, Murray E, Bradley SM, Hsu V, McGuire L and Schulz W (2015) ‘Non-modifiable worker and workplace risk factors contributing to workplace absence: A stakeholder-centred synthesis of systematic reviews’, Work, 52(2):353-73. doi:10.3233/WOR-152134.