Automated Prediction of Preservation Index in Library Environments through Multiple Regression Analysis

2.1 Data Collection

The study utilized empirical data collected over a three-month period at Brawijaya University Library. Environmental variables critical to preservation – temperature, humidity, and light exposure – were continuously recorded to capture the dynamic conditions within the library environment. This comprehensive dataset provides a solid foundation for analyzing the effects of environmental factors on material preservation (Camuffo 2019; Li and Tang 2013).

In this study, the environmental conditions within the library were meticulously monitored using an advanced Internet of Things (IoT) system. The system comprised DHT11 sensors known for their reliable measurement of temperature and humidity. These sensors were strategically placed in three distinct library rooms to ensure comprehensive data collection across varied conditions. A Raspberry Pi was employed as the central data processing unit, chosen for its ability to handle complex algorithms and manage large datasets effectively. This setup facilitated continuous data recording, with the Raspberry Pi programmed to collect readings from the sensors at one-minute intervals over a three-month period. The data was transmitted over a secured Wi-Fi network, allowing for real-time data access and remote monitoring by researchers. This frequent sampling rate provided a robust dataset, critical for analyzing the effects of environmental conditions on the library’s preservation index. Figure 1 below illustrates the complete configuration of the IoT system, showing the placement of sensors, their connection to the Raspberry Pi, and the subsequent data flow to the storage and analysis systems, thereby providing a visual representation of the systematic approach taken in capturing and processing the environmental data essential for this study.

Figure 1:

The IoT monitoring system.

Data was collected during operational hours, from 8:00 a.m. to 5:00 p.m. on working days, to ensure that the environmental conditions measured reflected the library spaces’ typical daily usage patterns. Each room was monitored sequentially over a one-month period to avoid data overlap and ensure precise attribution of environmental influences to specific areas. Measurements were captured at one-minute intervals, offering a high-resolution view of environmental fluctuations.

On certain occasions, the start of data logging devices was delayed due to operational constraints, resulting in start times slightly later than the scheduled 8:00 a.m. start time. These delays were mainly due to the manual setup needed each day to activate the monitoring systems, causing initial recordings to begin at varied times, such as 8:03 a.m. or 8:12 a.m. Despite these staggered starts, the recording consistently finished at the intended time of 5:00 p.m.

Efforts were made to document and account for all instances of delayed starts, ensuring that the analysis remained robust. This approach allowed us to maintain the integrity of the data collection process, with detailed logging of any deviations from the planned schedule. These deviations were minimal and did not significantly impact the overall dataset or the conclusions drawn from it.

Three distinct rooms within the Universitas Brawijaya Library were utilized, each with unique characteristics affecting the internal environmental conditions. The White Label Room featured 22 active windows providing ample natural light, supplemented by a combination of fluorescent and downlight spot lamps, designed to evenly distribute light without causing damage to materials. The Red Label Room, used primarily for specialized collections, included 14 active windows and utilized solely downlight spot lamps to enhance visibility while preserving the integrity of the materials. The Thesis Storage Room was equipped with 10 active windows and a similar lighting setup to ensure consistent light levels across all reading and storage areas.

In this study, we concentrated exclusively on monitoring and analyzing the internal environmental conditions of library environments, such as temperature, humidity, and light exposure inside different rooms. External environmental variables, including outdoor temperature and humidity, were not measured as they were outside the scope of our investigation. This methodological choice was deliberate to isolate the effects of internal microclimatic conditions on the Preservation Index of library materials, ensuring that our analysis directly reflects the impact of controlled internal factors. This focus is intended to enhance the applicability of our results to indoor environmental management practices for library preservation.

To better represent the spatial characteristics of the monitored rooms, Figure 2 presents the internal layout of the White Label Room, including its orientation to the cardinal directions and location of windows and doors. The reading areas are located in the center of the room, surrounded by shelves that are mostly aligned perpendicular to window positions.

Figure 2:

Spatial layout of the White Label Room.

As shown in Figure 3, the room receives significant daylight from the eastern and southern sides, which affects both the temperature and light exposure throughout the day. These spatial features were considered in the environmental analysis and model interpretation.

Figure 3:

Photograph of the White Label Room.

In our analysis, we combined data from three different rooms within the same library building – assuming that the environmental conditions are uniformly consistent across these spaces. This assumption is based on the fact that all rooms are situated in close proximity within the same building, under the same Heating, Ventilation, and Air Conditioning (HVAC) system, and have similar structural designs and exposure to environmental factors. Statistical analyses were conducted on the aggregated data to assess the overall impact of environmental conditions on the Preservation Index of library materials. This approach allows us to leverage a larger dataset and enhance the reliability of our findings by reducing variability due to small sample sizes in individual rooms.

An advanced Internet of Things (IoT) system was employed to monitor and record environmental conditions within the library accurately. This subsection details the setup and functionality of the system used.

2.2 Interpolation of Preservation Index Data

The Preservation Index (PI) values were derived from interval data provided in established preservation reference books. Interpolation techniques were applied to the interval PI data to create a continuous dataset suitable for statistical analysis. This process involved using methods like spline or linear interpolation to estimate PI values for each corresponding set of environmental measurements, thus filling gaps between the known values (Harim et al. 2020; Lai and Kaplan 2022; Renka 1987).

2.3 Regression Analysis

A regression model was developed to predict the PI based on the interpolated data and the measured environmental variables (Deepika and Channegowda 2020; Kareem et al 2022). The equation defines the model:

where β0, β1, β2, β3β0, β1, β2, and β3 are the coefficients estimated by the regression. The model was trained using a portion of the dataset and validated with the remaining data to assess its predictive accuracy and reliability.

2.4 Comparative Analysis

The effectiveness of each regression model in predicting PI was compared against the interpolated PI values from the book. Statistical measures such as mean absolute error (MAE) were used to quantify the accuracy and responsiveness of each method. This comparative analysis aims to identify the most reliable method for calculating the Preservation Index in library environments, considering the practical applicability of each method under real-world conditions.

3.1 Environmental Conditions in the Monitored Rooms

To better understand the microenvironmental context in which the preservation models were applied, this section presents trends in temperature, relative humidity, and light exposure observed in each room during the monitoring periods.

3.2 Temperature Trends

Figure 4 presents the daily average temperature trends observed in each monitored library room during separate data collection periods: White Label Room, Red Label Room, and Thesis Storage Room. Despite being measured in different months, the graphs are displayed using the same temporal format to enable direct visual comparison.

Figure 4:

Daily average temperature trends in each monitored library room.

The White Label Room exhibited relatively stable temperature fluctuations, while the Red Label Room showed a more pronounced daytime rise, likely due to higher solar exposure. The Thesis Storage Room exhibited the lowest overall temperature trend, possibly attributed to its shaded orientation and air circulation pattern.

These differences in thermal dynamics are important to consider when interpreting the regression results, as temperature variation has a direct impact on the Preservation Index.

3.3 Humidity Trends

Figure 5 presents the daily average humidity recorded in each of the monitored rooms during their respective data collection periods. The White Label Room showed consistently high humidity levels, ranging between 78 and 82 %, with minimal daily variation. The Red Label Room, in contrast, experienced more fluctuation, possibly due to greater air exchange from open windows or inconsistent airflow control. The Thesis Storage Room maintained moderate humidity levels around 60–65 %, indicating more stable environmental control or better insulation. These variations in humidity levels may influence the Preservation Index and are relevant when assessing the relative stability of each environment.

Figure 5:

Daily average relative humidity trends in each monitored library room.

3.4 Light Trends

Figure 6 illustrates the daily average light exposure levels recorded in the White Label, Red Label, and Thesis Storage Rooms. Among the three rooms, the White Label Room recorded the highest and most variable light exposure, reflecting its larger windows and minimal shading. The Red Label Room had consistently low light levels, which may be attributed to the use of curtains or internal room placement. The Thesis Storage Room exhibited moderate and relatively stable light exposure, suggesting a balanced combination of natural and artificial lighting.

Figure 6:

Daily average light exposure in the monitored rooms.

These environmental light differences are relevant for preservation considerations, as high and fluctuating exposure to light can accelerate fading, deterioration, and material aging, especially for paper-based archival collections.

3.5 Red Label Room

Figure 7 shows a Pearson correlation heatmap that helps us understand how temperature, humidity, and light exposure relate to each other in the Red Label Room of the library. The analysis reveals a strong negative relationship between temperature and humidity, with a correlation coefficient of −0.76. This means that when temperature goes up, humidity tends to go down significantly in this room. On the other hand, light exposure does not seem to have much connection with either temperature or humidity, showing very weak correlations of −0.10 and −0.09, respectively. This suggests that light changes happen independently of the area’s temperature or humidity changes.

Figure 7:

Pearson correlation for Red Label collection room.

The regression model developed for the Preservation Index (PI) in the Red Label Room is summarized by the equation:

where,

Y_RED represents the preservation index (PI), θ _RED is the temperature, η _RED is the humidity, and λ _RED is the light exposure in the Red Label Room. The intercept of 87.92 suggests a high baseline PI under optimal conditions where temperature, humidity, and light are at their reference levels, when all the independent variables are set to zero. The coefficients associated with temperature (θ _RED ) and humidity (η _RED ) are −0.7043 and −0.2826, respectively, indicating that increases in these variables lead to a decrease in the PI. Specifically, a one-unit increase in temperature is associated with a 0.7043 unit decrease in the PI, highlighting the sensitivity of preservation conditions to temperature fluctuations. Similarly, a one-unit increase in humidity results in a 0.2826 decrease in the PI, affirming the detrimental impact of higher humidity levels on material preservation.

The coefficient for light exposure (λ _RED ), at −0.0004 is significantly smaller, suggesting a minimal impact of light on the PI in this specific environment. This minimal influence reflects the relative insignificance of light exposure compared to temperature and humidity in affecting the preservation conditions within the Red Label Room.

3.6 White Label Room

Figure 8 presents the Pearson correlation heatmap for the White Label Room, providing insights into the interdependencies among temperature (°F), humidity (%), and light exposure. The heatmap indicates a moderate negative correlation of −0.42 between temperature and humidity, suggesting that humidity tends to decrease in this specific environment as temperature increases. This inverse relationship might be attributed to the air’s reduced capacity to hold moisture at higher temperatures, which is crucial for strategic preservation planning in archival environments.

Figure 8:

Pearson correlation for White Label collection room.

Conversely, the correlation between light exposure and both temperature and humidity are notably weak, with coefficients of −0.03 and 0.11, respectively. These values indicate that light exposure in the White Label Room varies almost independently of temperature and humidity changes. The near-zero correlation between temperature and light exposure (−0.03) implies that fluctuations in temperature do not significantly affect the light conditions within the room, and the positive but slight correlation (0.11) between humidity and light exposure suggests minimal co-variation.

The regression model tailored for assessing the Preservation Index (PI) within the White Label Room is encapsulated by the following equation:

where,

Y_WHITE denotes the PI, θ _WHITE represents the temperature, η _WHITE signifies the humidity, and λ _WHITE indicates light exposure in the White Label Room.

The model’s intercept at 91.75 suggests a relatively high baseline PI under controlled environmental conditions where temperature, humidity, and light are set at their baseline values. The temperature coefficient of −0.7978 indicates a significant negative impact on the PI; this implies that an increase in temperature by one unit decreases the PI by approximately 0.798 units, underscoring the critical influence of temperature regulation in the preservation environment.

Similarly, the humidity coefficient of −0.2338 suggests that increases in humidity also detrimentally affect the PI, albeit to a lesser extent than temperature. This coefficient reinforces the importance of maintaining optimal humidity levels to mitigate its negative impact on archival materials.

Interestingly, the coefficient for light exposure at −0.0002 is extremely small, suggesting that within the range of variation observed in this study, light exposure has an almost negligible direct impact on the PI. This minimal influence suggests that, for the White Label Room, other environmental controls might be prioritized over stringent light exposure management.

3.7 Thesis Storage Room

Figure 9 illustrates the Pearson correlation heatmap for environmental variables within the Thesis Storage Room. The analysis reveals a moderate negative correlation between temperature (°F) and humidity (%), with a coefficient of −0.39, indicating that increases in temperature are generally accompanied by decreases in humidity within this environment. This inverse relationship is crucial for understanding the storage conditions of theses, as temperature and humidity are known to significantly influence paper preservation.

Figure 9:

Pearson correlation for Thesis Storage Room.

In contrast, the correlation between temperature and light exposure is also negative, marked at −0.22, suggesting that higher temperatures slightly correlate with lower light levels in this room. However, this relationship is weaker compared to the temperature-humidity interaction. Light exposure shows a minimal positive correlation with humidity, recorded at 0.12, which is not substantial but indicates that slight increases in humidity could coincide with increases in light levels, although this effect is relatively negligible.

The regression model for the Thesis Storage Room, aimed at predicting the Preservation Index (PI), is succinctly formulated as follows:

in this model, Y_THESIS represents the PI, θ _THESIS denotes the temperature, η _THESIS denotes the humidity, and λ _THESIS indicates light exposure within the Thesis Storage Room.

The intercept value of 147.6966 suggests a relatively high baseline PI under controlled conditions, where all environmental variables are at their baseline or reference levels. This high baseline indicates that the preservation conditions are optimal when the environmental factors are maintained at standard levels, typical of controlled archival settings.

The coefficient for temperature (θ _THESIS ) at −1.2844 indicates a significant negative impact on the PI. This suggests that each unit increase in temperature results in a substantial decrease in the PI by approximately 1.2844 units, highlighting the critical need to manage temperature variations carefully within the storage area.

Similarly, the humidity’s coefficient (η _THESIS ) of −0.4971 reinforces the negative effect of increasing humidity on the PI. Although the impact is less severe than temperature, it underscores the importance of controlling humidity to prevent detrimental effects on the preservation of theses.

Lastly, the coefficient for light exposure (λ _THESIS ) at −0.0044, while still negative, indicates a very minor influence on the PI. This minimal impact suggests that light exposure does not play a significant role in the degradation of materials in the Thesis Storage Room, compared to temperature and humidity.

3.8 Environmental Insights Across Rooms

This section presents the collection and analysis of environmental data from multiple storage areas within the library, encompassing the White, Red, and Thesis Storage Rooms.

The consolidated dataset offers a comprehensive view of the interactions among environmental variables – temperature, humidity, and light exposure – across these diverse yet interconnected spaces. By examining this data, we seek to identify underlying patterns that potentially influence material preservation across different storage environments.

The Pearson correlation heatmap shown in Figure 10 assesses the interactions among environmental variables – temperature (suhu_f), humidity (kelembaban_%), and light exposure (cahaya) – in the combined settings of the White, Red, and Thesis Storage Rooms. This visualization indicates generally weak correlations, suggesting that these variables operate independently across the different environments. Specifically, there is a slight negative correlation of −0.09 between temperature and humidity, showing that higher temperatures do not strongly correlate with lower humidity levels. Additionally, the correlation between temperature and light exposure is also weak at −0.03, indicating minimal dependence of light levels on temperature changes. Notably, light exposure has a small positive correlation with humidity at 0.12, implying that increases in humidity might slightly enhance light exposure, although this link is too weak to infer a strong dependency.

Figure 10:

Pearson correlation across rooms.

The regression model developed to predict the Preservation Index (PI) across all room settings – White, Red, and Thesis Storage Rooms – is expressed by the equation:

here, Y_ALL represents the PI, θ _ALL denotes temperature, η _ALL signifies humidity, and λ _ALL indicates light exposure.

The intercept value of 123.3949 suggests a high baseline PI, assuming that temperature, humidity, and light are at their reference levels. This implies that under optimal conditions, the preservation environment is highly favourable. The coefficient for temperature (θ _ALL ) at −1.0472 indicates a substantial negative impact on the PI; specifically, a one-unit increase in temperature correlates with a decrease in the PI by approximately 1.0472 units, underscoring the critical sensitivity of the preservation conditions to temperature changes.

Furthermore, the coefficient for humidity (η _ALL ) at −0.3882 also demonstrates a negative effect on the PI, albeit less pronounced than temperature, yet still significant for material preservation strategies.

Lastly, the coefficient for light exposure (λ _ALL ) at −0.0042, while negative, indicates a relatively minor impact on the PI compared to temperature and humidity. This minimal coefficient suggests that light exposure, within the observed ranges, does not significantly influence the preservation conditions as much as the other environmental variables.

3.9 Statistical Assessment of Regression Models

We explore the outcomes of the regression analysis by focusing on statistical measures that assess the performance and validity of our models across different environmental settings. These include R² (R-squared), the F-statistics, and the Durbin-Watson statistics. R² provides a measure of how much of the variability in the dependent variable can be explained by the independent variables in the model. A higher R² value indicates a better fit of the model to the data, suggesting that the model explains a large proportion of the variability. The F-statistics evaluate the overall significance of the regression model, testing whether the observed relationships between the dependent and independent variables are statistically significant. A higher F-statistic value signals that the model is statistically meaningful, which implies that the predictors collectively have a significant effect on the outcome variable. Meanwhile, the Durbin-Watson statistic helps detect the presence of autocorrelation in the residuals of the regression model. Values close to 2 indicate no autocorrelation, affirming that the residuals are independent and that the regression model is appropriately specified.

Table 1 provides a comparison of regression models for various environmental settings within the library, specifically the White, Red, and Thesis Storage Rooms, along with a combined model. The R² values across these models are impressively high, indicating that nearly all the variability in the Preservation Index can be explained by the models for each respective environment. Specifically, the combined room model has the highest R² value at 0.984, suggesting the most robust fit among the models.

Looking at the F-statistics, which test the overall significance of each regression model, we observe a clear hierarchy in model significance, though all models demonstrate statistically significant relationships between environmental factors and the Preservation Index. The Red Room model exhibits the highest F-statistic at 634,700, indicating an exceptionally strong predictive relationship, likely due to specific environmental sensitivities unique to this room. The Combined Rooms model follows with an F-statistic of 206,200, underscoring its robustness across multiple environments and validating its utility for overarching environmental strategies. The White Room, with an F-statistic of 141,400, and the Thesis Storage Room, at 57,380, still present significant models that effectively capture environmental impacts on preservation. These statistics not only affirm the validity of each model but also highlight the varying degrees of environmental influence across different room settings, suggesting that while some areas may be more sensitive to changes, all areas are significantly affected by their respective environmental conditions.

The Durbin-Watson statistic values for each environment demonstrate that the residuals from the regression models exhibit minimal to no autocorrelation, ensuring the independence of observations. This aspect is crucial for validating the reliability and accuracy of our predictive models. Specifically, the values – 0.015 for the Red Room, 0.023 for the White Room, 0.019 for the Thesis Storage Room, and 0.017 for the Combined Rooms – are all close to 2, indicating that autocorrelation does not significantly affect the outcomes of these models. Despite being slightly below the ideal value of 2, these statistics are within an acceptable range, confirming that the residuals are appropriately random. This evidence supports the robustness of our environmental models across different room settings, suggesting that they provide unbiased and consistent estimations of how environmental variables influence the Preservation Index.

Overall, the analysis of regression models across various library environments highlights that all models are statistically significant and reliable, with each model effectively capturing the influence of environmental variables on the Preservation Index. The Red Room model stands out as particularly strong, evidenced by its highest F-statistic and close-to-ideal Durbin-Watson value, suggesting that it may offer the most accurate predictions within its specific setting. The Combined Rooms model also demonstrates a high level of robustness, making it an excellent candidate for overarching environmental strategies, especially for applications that require a unified approach across different environments. Although the White Room and Thesis Storage Room models have slightly lower metrics, they are still within acceptable ranges and provide valuable insights into their respective settings. Given these findings, all models are deemed reliable, but the Red Room and Combined Rooms models show the most promise for further application.

Moving forward, we will assess the predictive power of these regression models by testing them against a separate dataset designated as the test data. By calculating the Mean Absolute Percentage Error (MAPE) for each model, we will better understand their practical applicability and accuracy. The following subsection will detail this process and the results obtained, providing further insights into the efficacy of these environmental models.

3.10 MAPE Analysis for Room-Specific Regression Models

In this section, we evaluate the predictive accuracy of the regression models for different library environments by comparing their Mean Absolute Percentage Error (MAPE). MAPE is a commonly used measure to assess the accuracy of forecasting models. It expresses the average absolute difference between the predicted values and the actual values as a percentage of the actual values. The formula for calculating MAPE is as follows:

where A_i represents the actual value, F_i represents the forecasted value, and n is the number of observations. A lower MAPE value indicates a more accurate model, as it suggests that the predictions are closer to the actual values. By comparing the MAPE values across the different room-specific models, we can determine which model offers the highest predictive accuracy. This analysis will provide critical insights into the practical applicability of each model for environmental management within the library (Table 2).

The MAPE results reveal varying levels of predictive accuracy across the different room models. The Red Room model stands out with the lowest MAPE of 0.4921, indicating it provides the most accurate predictions among all the models. This suggests that the environmental conditions in the Red Room are well understood and effectively modeled, making it a reliable choice for precise environmental management. The White Room and Thesis Storage Room models have slightly higher MAPEs, at 1.1106 and 1.2440 respectively, but these are still within acceptable ranges for practical use. These results suggest that while these models are fairly accurate, there may be opportunities for refinement to improve their predictive capabilities. The Combined Room model, with a MAPE of 2.6515, shows the least accuracy, likely due to the complexity of integrating data from multiple environments. However, the Combined Room model remains a viable option, as it has been designed to account for the varying conditions of temperature, humidity, and light across all rooms. Despite its higher MAPE, this model still achieves a relatively high level of accuracy, making it suitable for general environmental management applications where a broad overview is required. Therefore, while the Red Room model is recommended for precise monitoring, the Combined Room model is also a strong candidate for situations where comprehensive, multi-environment analysis is needed.