Mean Average Percentage Error Research Articles

Lifespan prediction of lithium-ion batteries based on various extracted features and gradient boosting regression tree model

Accurate battery lifespan prediction is critical for the quality evaluation and long-term planning of battery management systems. As battery degradation process is typically nonlinear, accurate early prediction of cycle life with significantly less degradation is extremely challenging. Approaches using machine learning techniques, which are mechanism-agnostic alternatives, to predict battery lifespan are therefore becoming more and more attractive. In this paper, a gradient boosting regression tree (GBRT) is proposed to model complex nonlinear battery dynamics and predict battery lifespan through various extracted battery features. Essentially, the GBRT works by constructing additional trees through minimizing the prediction residues from existing base models. It can identify complex and nonlinear end-to-end relationship and meanwhile provide relative importance for each input feature. Various potential features including voltage-related features, capacity-related features, and temperature-related features are constructed in both discharge time dimension and life cycle dimension and explored for effective lifespan prediction. Key hyper-parameters including learning rate, number of trees, and maximum number of splits are investigated for optimal GBRT prediction. Comparative studies confirm that the proposed method is significantly superior to other machine learning algorithms for battery lifespan prediction with limited data samples and various input features, with mean average percentage error around 7%.

Sales forecasting in rapid market changes using a minimum description length neural network

In this study, we address a real sales forecasting problem of a multinational fashion retailer. The fashion retailer has been operating retail stores and warehouses in a major Asian city for decades. The volume of daily sales of each stock keeping unit in this city needs to be determined to develop an effective and efficient inventory and logistics system for the retailer. Based on complicated real sales data sets, we use a minimum description length neural network (MDL-NN) which searches for the optimal model size to predict this value. In order to address the problem of intermittent sales data and zero actual sales, we propose a revised mean absolute percentage error (RMAPE) measurement for performance evaluation. Our experimental results show that for most test data sets, the evaluation results based on RMAPE are consistent with those based on two other established measurements, the symmetric mean average percentage error and mean absolute scaled error. Specifically, on all three performance indicators, the MDL-NN method achieves a smaller error than almost all other commonly used sales forecasting methods on almost all test data sets. Finally, we examine the gap between the performance indicators between the forecasting result values in decimal format, which are directly generated by the forecasting methods, and the forecasting result values in integer format, which are rounded off from the decimal numbers. Our findings indicate that this gap is not trivial and should not be ignored.

The design of a smart home controller based on ADALINE

This paper proposes a prototype of an improved smart home controller that implements a neural network-based algorithm for enabling the controller to make decisions and act based on the current condition. Unlike previous approaches, this design also utilizes the use of IoT (internet of thing) technology and neural network based-algorithm for developing the controller. Since a smart home is equipped with various sensors, actuators, smart appliances, and mobile terminals, all of these devices need to be connected to the Internet to be able to communicate and provide services for its occupants. The construction of the proposed controller is carried out through several procedures, i.e. the implementation of the ADALINE (adaptive linear) as the neural network method, the design of the smart home controller prototype, and the validation process using mean average percentage error (MAPE) calculation. This prototype integrates functionalities of several household appliances into one application controlled by a smartphone. ADALINE is applied as an algorithm to predict output when the controller is in automatic mode. Although the obtained accuracy value is still not satisfactory, the value is bound to change when testing on more data. The work published in this paper may encourage the implementation of smart technology in more households in Indonesia.

Comparative Study on Fuzzy Models for Crop Production Forecasting

Fuzzy sets theory is a very useful technique to increase effectiveness and efficiency of forecasting. The conventional time series is not applicable when the variable of time series are words variables i.e. variables with linguistic terms. As India and most of the Asian countries are of agriculture-based economy with very smaller farmer land holding area in comparison to America, Australia and Europe counterparts, it becomes more important for these countries to have an approximate idea regarding future crop production. It not only will help in planning policies for future but also will be a great help for farmers and agro based companies for their future managements. For small area production, soft computing technique is an important and effective tool for predicting production, as agriculture production involve a high degree of uncertainties in many parameters. In the present study, 21 years agricultural crop yield data is used and a comparative analysis of forecast is done with three fuzzy models. The robustness of the model is tested on real time agricultural farm production data of wheat crop of G.B. Pant University of Agriculture and Technology Pantnagar, India. As soft computing techniques involve uncertainty of the system under study, it becomes more and more important for forecasting models to be accurate with the prediction. The efficiency of the three models is examined on the basis of statistical errors. The models under study are judged on the basis of Mean Square Error and average percentage error. The results of the study are in case of small area production prediction and will encourage for predicting large scale production.

Assessment of stacked unidirectional and bidirectional long short-term memory networks for electricity load forecasting

Electricity load forecasting has been a substantial problem in the electric power system management process. An accurate forecasting model is essential to avoid imprecise predictions that can negatively affect system efficiency, economy, and sustainability. Among several prediction techniques, deep learning methods, especially the Long Short-Term Memory (LSTM), have been shown to have a superior performance in predicting the electricity load consumption. However, the consequences of using these methods have not fully been explored in terms of the various hidden layer structures, the depth of the model architecture, and the impact of tuning the model hyperparameters. In this paper, a systematic experimental methodology has been conducted to investigate the impact of using deep-stacked unidirectional (Uni-LSTM) and bidirectional (Bi-LSTM) networks on predicting electricity load consumption. In particular, two stacked configurations, which include two and three LSTM layers, are compared with the single-layered LSTM for both types to show the significant importance of adding the stacked layers. Moreover, for each proposed configuration, a hyperparameter optimization tool has been implemented to obtain the best model. The results indicate that the deep-stacked LSTM layers have no significant improvement in the prediction accuracy; nevertheless, they consume almost twice the time of the single-layered models. Also, the Bi-LSTM networks outperform the Uni-LSTM networks by 76.25%, 75.49%, and 75.35% in terms of Root Mean Square Error (RMSE), with respect to one, two, and three-layer model configurations, respectively. Furthermore, regarding the prediction accuracy comparison over the total tested period, the optimized Bi-LSTM model outperforms both the optimized Uni-LSTM model by 75.98%, 89.1%, and 89.37%, and the Support Vector Regression (SVR) model by 82.54%, 92.59%, and 92.89% in terms of (RMSE), the Mean Average Percentage Error (MAPE), and Mean Absolute Errors (MAE).

Prediction and Simulation Spatio-Temporal Support Vector Regression for Nonlinear Data

<p>Spatio-temporal model forecasting method is a forecasting model that combines forecasting with a function of time and space. This method is expected to be able to answer the challenge to produce more accurate and representative forecasting. Using the ability of method Support Vector Regression in dealing with data that is mostly patterned non-linear premises n adding a spatial element in the model of forecasting in the form of a model forecasting Spatio- Temporal. Some simulations have done with generating data that follows the Threshold Autoregressive model. The models are correlated into spatial points generated by several sampling methods. Simulation models are generated to comparing the accuracy between model Spatio-Temporal Support Vector Regression and model ARIMA based on Mean Error, Mean Average Error, Root Mean Square Error, and Mean Average Percentage Error. Based on the evaluation results, it is shown that forecasting with the Spatio-Temporal Support Vector Regression model has better accuracy than forecasting ARIMA.</p>

Short-Term Wind Speed Forecasting Using Statistical and Machine Learning Methods

Wind offers an environmentally sustainable energy resource that has seen increasing global adoption in recent years. However, its intermittent, unstable and stochastic nature hampers its representation among other renewable energy sources. This work addresses the forecasting of wind speed, a primary input needed for wind energy generation, using data obtained from the South African Wind Atlas Project. Forecasting is carried out on a two days ahead time horizon. We investigate the predictive performance of artificial neural networks (ANN) trained with Bayesian regularisation, decision trees based stochastic gradient boosting (SGB) and generalised additive models (GAMs). The results of the comparative analysis suggest that ANN displays superior predictive performance based on root mean square error (RMSE). In contrast, SGB shows outperformance in terms of mean average error (MAE) and the related mean average percentage error (MAPE). A further comparison of two forecast combination methods involving the linear and additive quantile regression averaging show the latter forecast combination method as yielding lower prediction accuracy. The additive quantile regression averaging based prediction intervals also show outperformance in terms of validity, reliability, quality and accuracy. Interval combination methods show the median method as better than its pure average counterpart. Point forecasts combination and interval forecasting methods are found to improve forecast performance.

Using an Accelerometer-Based Step Counter in Post-Stroke Patients: Validation of a Low-Cost Tool.

Monitoring the real-life mobility of stroke patients could be extremely useful for clinicians. Step counters are a widely accessible, portable, and cheap technology that can be used to monitor patients in different environments. The aim of this study was to validate a low-cost commercial tri-axial accelerometer-based step counter for stroke patients and to determine the best positioning of the step counter (wrists, ankles, and waist). Ten healthy subjects and 43 post-stroke patients were enrolled and performed four validated clinical tests (10 m, 50 m, and 6 min walking tests and timed up and go tests) while wearing five step counters in different positions while a trained operator counted the number of steps executed in each test manually. Data from step counters and those collected manually were compared using the intraclass coefficient correlation and mean average percentage error. The Bland–Altman plot was also used to describe agreement between the two quantitative measurements (step counter vs. manual counting). During walking tests in healthy subjects, the best reliability was found for lower limbs and waist placement (intraclass coefficient correlations (ICCs) from 0.46 to 0.99), and weak reliability was observed for upper limb placement in every test (ICCs from 0.06 to 0.38). On the contrary, in post-stroke patients, moderate reliability was found only for the lower limbs in the 6 min walking test (healthy ankle ICC: 0.69; pathological ankle ICC: 0.70). Furthermore, the Bland–Altman plot highlighted large average discrepancies between methods for the pathological group. However, while the step counter was not able to reliably determine steps for slow patients, when applied to the healthy ankle of patients who walked faster than 0.8 m/s, it counted steps with excellent precision, similar to that seen in the healthy subjects (ICCs from 0.36 to 0.99). These findings show that a low-cost accelerometer-based step counter could be useful for measuring mobility in select high-performance patients and could be used in clinical and real-world settings.

An artificial neural network approach for the prediction of dynamic viscosity of MXene-palm oil nanofluid using experimental data

Given the excellent thermal properties of MXene, MXene nanomaterials-based nanofluids may have the potential of being used as heat transfer fluids. In this work, we have investigated the dynamic viscosity of MXene-palm oil nanofluid. To prepare the nanofluid, MXene ( $$\hbox {Ti}_{{3}}\hbox {C}_{{2}}$$ ) nanoflakes were first synthesized by the wet chemistry method. Then, nanofluids with 0.01, 0.03, 0.05, 0.08, 0.1, and 0.2 mass% concentrations of MXene nanoflakes were prepared, and their dynamic viscosity values were determined for a wide temperature range of 18–100 $$^\circ \hbox {C}$$ . The dynamic viscosity of the developed nanofluid, for each concentration, was found to be a strong function of temperature and decreased with increasing temperature. The effect of MXene nanoflakes concentration on the dynamic viscosity was found to be more significant at lower temperatures than at higher temperatures. A multilayer perceptron artificial neural network model was also developed in the study. To develop the model, temperature and nanoflakes concentration were given as inputs to the ANN model, whereas dynamic viscosity was the output of the model. The optimum model was found through the trial and error method. Statistical parameters such as mean square error, mean average percentage error, and correlation coefficient (R) were used to evaluate the performance of the developed model. The values of these parameters, for the optimum ANN model, were found to be 4.733E−05, 0.507%, and 0.99975, respectively. 95.67% of the deviations were also found to be in range of ± 2%. The developed model showed good performance, and its predictions were in excellent agreement with the experimental data.

Modeling of mixed convection in an enclosure using multiple regression, artificial neural network, and adaptive neuro-fuzzy interface system models

In this study, the heat transfer characteristics of laminar combined forced convection through a horizontal duct are obtained with the help of the numerical methods. The effect of the geometrical parameters of the cavity and Reynolds number on the heat transfer is investigated. New heat transfer correlation for hydrodynamically fully developed, laminar combined forced convection through a horizontal duct is proposed with an average error of 6.98% and R2 of 0.8625. The obtained correlation results are compared with the artificial neural network and adaptive neuro-fuzzy interface system models. Due to the obtained results, good agreement is identified between the numerical results and predicted adaptive neuro-fuzzy interface system results. In conclusion, it is seen that adaptive neuro-fuzzy interface system can predict the Nusselt number distribution with a higher accuracy than the developed correlation and the artificial neural network model. The developed adaptive neuro-fuzzy interface system model predicts the Nusselt number with 1.07% mean average percentage error and 0.9983 R2 value. The effect of the different training algorithms and their ability to predict Nusselt number distribution are examined. According to the results, the Bayesian regulation algorithm gives the best approach with a 2.235% error. According to the examination that is performed in this study, the adaptive neuro-fuzzy interface system is a powerful, robust tool that can be used with confidence for predicting the thermal performance.

Comparison of R404A condensation heat transfer and pressure drop with low global warming potential replacement candidates R448A and R452A

The objective of this paper was to measure and compare the heat transfer coefficient and pressure drop of R404A and two potential replacement candidate refrigerants with a lower global warming potential, R448A and R452A. Experiments were conducted inside a 4.7 mm horizontal tube at mass fluxes ranging from 100 to 800 kg m−2 s−1, and at three different saturation conditions (40, 50, and 60 ∘C). R448A and R452A are zeotropic refrigerant mixtures, with temperature glides of approximately 4 ∘C and 3 ∘C, respectively. The Cavallini et al. (2006) correlation predicted the heat transfer data best for all three refrigerants, once mixture effects were accounted for using the equilibrium Silver (1947), Bell and Ghaly (1973) method. The Haraguchi et al. (1994) correlation predicted the R484A frictional pressure drop data best (mean average percent error, MAPE = 15%), while the Friedel (1979) correlation predicted the R404A (MAPE = 15%) and R452A data (MAPE = 14%) pressure drop the best. The heat transfer coefficients of R404A were slightly higher than those of R452A and slightly lower than those of R448A at equivalent conditions. At the same mass flux, (G = 600 kg m−2 s−1), R448A and R452A pressure drops are on average 81% and 3.3% higher than that of R404A. For the same cooling capacity and condenser design, the pressure drop of R404A and R452A are approximately the same and the pressure drop of R448A is approximately 16% lower than that of R404A.

ANN modelling of CO2 refrigerant cooling system COP in a smart warehouse

Industrial cooling systems consume large quantities of energy with highly variable power demand. To reduce environmental impact and overall energy consumption, and to stabilize the power requirements, it is recommended to recover surplus heat, store energy, and integrate renewable energy production. To control these operations continuously in a complex energy system, an intelligent energy management system can be employed using operational data and machine learning. In this work, we have developed an artificial neural network based technique for modelling operational CO2 refrigerant based industrial cooling systems for embedding in an overall energy management system. The operating temperature and pressure measurements, as well as the operating frequency of compressors, are used in developing operational model of the cooling system, which outputs electrical consumption and refrigerant mass flow without the need for additional physical measurements. The presented model is superior to a generalized theoretical model, as it learns from data that includes individual compressor type characteristics. The results show that the presented approach is relatively precise with a Mean Average Percentage Error (MAPE) as low as 5%, using low resolution and asynchronous data from a case study system. The developed model is also tested in a laboratory setting, where MAPE is shown to be as low as 1.8%.

Univariate data-driven models for glucose level prediction of CGM sensor dataset for T1DM management

The advent of machine learning has made a remarkable impact in the field of healthcare. Diabetes mellitus is a metabolism abnormality that is posing severe threat, exercising substantial pressure on human health worldwide. Diabetes mellitus is a public health problem around the world. In 1980, 108 million adults worldwide had diabetes. By 2040 the number is expected to reach 642 million adults. Hence extensive research in interdisciplinary field that uses skills from various fields such as statistics machine learning, artificial intelligence, visualization, etc. is carried out for better management of diabetes. In this paper, the focus is to use time series forecasting algorithms. Data-driven models in time series machine learning are used to derive meaningful and appropriate information from large volumes of blood glucose level and related data for precise forecasting of upcoming blood glucose level fluctuations. Not only can the patient and physician be informed beforehand, to avert complications, but also it aids in predicting response to certain medications with ease. In this case, univariate data-driven models from time series machine learning algorithms are implemented on 2 different continuous glucose monitoring sensor datasets: Libre Pro dataset of 10 patients and Ohio T1DM dataset of 6 patients. A comparison of performance evaluation metrics of the different time series machine learning algorithms is drawn based on root mean squared error (RMSE), mean average percentage error (MAPE) and Theil’s U, which are statistical analyses, and Clarke’s error grid, which is clinical analysis for prediction horizon from 15 to 45 min. Using Holt’s Linear AAN Algorithm on Libre Pro dataset with alpha and beta of 0.99 provided the least error among exponential smoothing algorithms with RMSE of 7.98 mg/dl for 15 min, 19.47 mg/dl for 30 min and 28.40 mg/dl for 45 min prediction horizon. Theil’s U coefficient was 0.12 for 15 min, 0.39 for 30 min and 0.72 for 45 min prediction horizon. Autoregressive Integrated Moving Average (ARIMA) Algorithm gave the best performance evaluation results with RMSE of 7.07 mg/dl for 15 min with a MAPE of 3.98. The performance results were on par when these algorithms were tested on Ohio T1DM dataset. ARIMA Algorithm gave the best performance evaluation results with RMSE of 13.14 mg/dl for 15 min with a MAPE of 8.213. The difference in the error coefficient for Ohio dataset was due to missing data.

Estimating Daily Global Solar Radiation with No Meteorological Data in Poland

The aim of the study was to calibrate coefficients and evaluate performance of simple, day-of-the-year, global solar radiation (H) models nominated from the literature. Day-of-the-year models enable estimation of global solar radiation when no meteorological data is available. The study used 16-year-long data series of daily H, taken at 15 actinometric stations located in various parts of Poland. The goodness-of-fit of the models to the actual long-term monthly average daily global solar radiation data expressed by determination coefficient (R2) ranges from 0.94 to 0.97. Depending on statistical indicators analysis (root mean square error—RMSE, mean absolute bias error—MABE, mean average percentage error—MAPE) the best model was selected. The averaged values of H computed by the recommended model deviate from those measured by 4.16% to 8.71%. Locally calibrated, day-of-the-year model provides satisfactory accuracy and—where meteorological data is unavailable—can be used to estimate mean monthly daily global solar radiation in Poland and similar climate conditions.

A HYBRID LEAST SQUARES SUPPORT VECTOR MACHINE WITH BAT AND CUCKOO SEARCH ALGORITHMS FOR TIME SERIES FORECASTING

Least Squares Support Vector Machine (LSSVM) has been known to be one of the effective forecasting models. However, its operation relies on two important parameters (regularization and kernel). Pre-determining the values of parameters will affect the results of the forecasting model; hence, to find the optimal value of these parameters, this study investigates the adaptation of Bat and Cuckoo Search algorithms to optimize LSSVM parameters. Even though Cuckoo Search has been proven to be able to solve global optimization in various areas, the algorithm leads to a slow convergence rate when the step size is large. Hence, to enhance the search ability of Cuckoo Search, it is integrated with Bat algorithm that offers a balanced search between global and local. Evaluation was performed separately to further analyze the strength of Bat and Cuckoo Search to optimize LSSVM parameters. Five evaluation metrics were utilized; mean average percent error (MAPE), accuracy, symmetric mean absolute percent error (SMAPE), root mean square percent error (RMSPE) and fitness value. Experimental results on diabetes forecasting demonstrated that the proposed BAT-LSSVM and CUCKOO-LSSVM generated lower MAPE and SMAPE, at the same time produced higher accuracy and fitness value compared to particle swarm optimization (PSO)-LSSVM and a non-optimized LSSVM. Following the success, this study has integrated the two algorithms to better optimize the LSSVM. The newly proposed forecasting algorithm, termed as CUCKOO-BAT-LSSVM, produces better forecasting in terms of MAPE, accuracy and RMSPE. Such an outcome provides an alternative model to be used in facilitating decision-making in forecasting.

A Hybrid Deep Learning-Based Traffic Forecasting Approach Integrating Adjacency Filtering and Frequency Decomposition

Traffic forecasting in urban area has attracted substantial attention in recent years due to its significant assitance for traffic dispatching and trip planning. However, this task is very challenging due to the complex dependencies inherent in the traffic process. In our study, recent and periodic dependencies are identified and used to describe the corresponding near-term and long-term effects of historical traffic data on future traffic states. Subsequently, during the recent dependency modeling, each road is found to correlate with its adjacent roads through traffic flow diffusion, then the correlation intensities are quantified and used to choose strongly correlated roads to build a critical road sequence. While for the periodic dependency modeling, since the historical speed series of a road segment exhibits multi-frequency attributes (i.e., low-frequency daily period and high-frequency stochastic fluctuation), wavelet transform is conducted to decompose the original speed series into low and high-frequency sub-series. On these bases, we propose a hybrid traffic speed forecasting model, flow and wavelet-integrated spatio-temporal network (FW-STN). In the FW-STN, the recent features are captured by the convolutional neural network (CNN) with near-term traffic data from the derived critical road sequence, and the periodic features are captured by the long short-term memory (LSTM) with the low and high-frequency sub-series. Both the recent and periodic features are then fused to conduct the final prediction. Experimental results on real traffic data show that the proposed approach outperforms nine state-of-the-art methods (with improvements of 3% ~15% in mean average percentage error).

Forecasting cesarean deliveries with robust time series models in a tertiary care hospital

Introduction: Caesarian section rate was 47% at Patan Hospital in 2014 despite the recommendation of keeping it below 15%. This has become a public health problem and it is debated as human right violation of childbearing women. This study aims use robust time series model to forecast caesarian deliveries to keep track of it at the hospital.
 Method: Univariate time series models were used to forecast 3-year caesarean deliveries at Patan Hospital using 60-month (2010-2014) data. A robust time series model with low mean average percentage error from validation period and without autocorrelation problem was selected and used to forecast caesarean deliveries for 2015-2017 period. 
 Result: Winter’s additive model had lowest validation forecasting error and showed decreasing trend of caesarian deliveries but it showed autocorrelation. Quadratic regression gave similar result but is also autocorrelation problem. Artificial Neural Network – Multilayer Perceptron model gave close forecasts but autocorrelation was not assessed. Best Autoregressive Integrated Moving Average (ARIMA) model gave valid forecasts without autocorrelation problem.
 Conclusion: ARIMA (0,1,1),(0,0,0) model with one difference and one level of moving error correction gave valid forecasts for Patan Hospital. Advanced univariate and multivariate time series models with large samples can be used to get precise forecasts of caesarean deliveries in Nepal.

Comparison of Holt and Brown's Double Exponential Smoothing Methods in The Forecast of Moving Price for Mutual Funds

Mutual funds are one of the promising investment media where the risk is directly proportional to the size of investment growth. With proper forecasting of NAV price movements will greatly help investors to make purchases and sales transactions, therefore the authors offer the use of two different forecasting methods namely Brown's method and Holt method in double exponential smoothing to get predictions of NAV price movements. The effectiveness of the use of the method will be measured from the value of Mean Average Percentage Error (MAPE). From the calculation results obtained by the data that the Holt method produces forecasting for 1809,657 with the best α value of 0.6 and MAPE of 0.644373568, while for the Holt method obtained forecasting value of 1810,924 with the α value and the best β value of 0.9 and 0.1 and the smaller MAPE value of 0.61604262 . Looking at the amount of MAPE generated, the Holt method has a smaller forecasting error rate when compared to Brown’s method.

Chest Movement and Respiratory Volume both Contribute to Thoracic Bioimpedance during Loaded Breathing

Bioimpedance has been widely studied as alternative to respiratory monitoring methods because of its linear relationship with respiratory volume during normal breathing. However, other body tissues and fluids contribute to the bioimpedance measurement. The objective of this study is to investigate the relevance of chest movement in thoracic bioimpedance contributions to evaluate the applicability of bioimpedance for respiratory monitoring. We measured airflow, bioimpedance at four electrode configurations and thoracic accelerometer data in 10 healthy subjects during inspiratory loading. This protocol permitted us to study the contributions during different levels of inspiratory muscle activity. We used chest movement and volume signals to characterize the bioimpedance signal using linear mixed-effect models and neural networks for each subject and level of muscle activity. The performance was evaluated using the Mean Average Percentage Errors for each respiratory cycle. The lowest errors corresponded to the combination of chest movement and volume for both linear models and neural networks. Particularly, neural networks presented lower errors (median below 4.29%). At high levels of muscle activity, the differences in model performance indicated an increased contribution of chest movement to the bioimpedance signal. Accordingly, chest movement contributed substantially to bioimpedance measurement and more notably at high muscle activity levels.

Forecasting emergency admissions due to respiratory diseases in high variability scenarios using time series: A case study in Chile

Respiratory diseases are ranked in the top ten group of the most frequent illness in the globe. Emergency admissions are proof of this issue, especially in the winter season. For this study, the city of Santiago de Chile was chosen because of the high variability of the time series for admissions, the quality of data collected in the governmental repository DEIS (selected period: 2014–2018), and the poor ventilation conditions of the city, which in winter contributes to increase the pollution level, and therefore, respiratory emergency admissions. Different forecasting models were reviewed using the Akaike Information Criteria (AIC) with other error estimators, such as the Root Mean Square Error (RMSE), for selecting the best approach. At the end, Seasonal Autoregressive Integrated Moving Average (SARIMA) model, with parameters (p,d,q)(P,D,Q)s=(2,1,3)(3,0,2)7, was selected. The Mean Average Percentage Error (MAPE) for this model was 7.81%. After selection, an investigation of its performance was made using a cross-validation through a rolling window analysis, forecasting up to 30 days ahead (testing period of one year). The results showed that error do not exceed a MAPE of 20%. This allows taking better resource managing decisions in real scenarios: reactive staff hiring is avoided given the reduction of uncertainty for the medium term forecast, which translates into lower costs. Finally, a methodology for the selection of forecasting models is proposed, which includes other constraints from resource management, as well as the different impacts for social well-being.