Adaptive Neuro-Fuzzy Inference System for Waste Prediction

ABSTRACT


I. Introduction
The ever-increasing volume of waste produced due to human activity poses a massive threat to both the environment and public health. Waste management has emerged as one of the most pressing concerns for governments and communities worldwide due to the ongoing rise in the global population and the acceleration of urbanization [1]. In recent years, there has been an increasing awareness of the detrimental effects of waste on the environment, such as contamination of the air and water [2], deterioration of the land [3], and emissions of greenhouse gases [4]. Ineffective waste management procedures have also been linked to various health issues, such as respiratory disorders, infectious infections, and malignancies [5]. As a direct consequence of this, waste management has risen to the top of the agenda for decision-makers in government, those in charge of waste management, and other stakeholders [6]. They are actively looking for ways to manage waste sustainably, cut down on the amount of waste generated, and lessen the adverse effects of waste on the environment and public health. The difficulty of accurately predicting the amount of waste produced in a particular location is one of the obstacles faced in waste management.
Prediction is a process that involves certain behaviors or phenomena that will occur in the future. Predictions can be made quantitatively or qualitatively. The quantitative measurement uses statistical methods, while the qualitative measurement is based on the opinion (judgment) of those who make predictions. Based on the time horizon, predictions can be grouped into three parts: long-term, medium-term, and short-term [7]. Predictions can be qualitative (not in the form of numbers) or quantitative (in the form of numbers). Qualitative predictions are difficult to do to obtain good results because the variables are very relative. Quantitative prediction is divided into two, namely: single prediction (point prediction) and interval prediction (interval prediction). A single prediction consists of one value, while an interval prediction consists of several values in the form of an interval (interval) bounded by lower limit values (lower limit prediction) and upper limit (high prediction) [8]. Prediction of the volume of landfill waste is carried out to assist related parties in making policies on the growth The volume of landfills that are increasingly piled up and not handled properly will have a negative impact, such as a decrease in public health. Therefore, predicting the volume of landfills with a high degree of accuracy is needed as a reference for government agencies and the community in making future policies. This study aims to analyze the accuracy of the Adaptive Neuro-Fuzzy Inference System (ANFIS) method. The prediction results' accuracy level is measured by the value of the Mean Absolute Percentage Error (MAPE). The final results of this study were obtained from the best MAPE test results. The best predictive results for the ANFIS method were obtained by MAPE of 3.36% with a data ratio of 6:1 in the North Samarinda District. The study results show that the ANFIS algorithm can be used as an alternative forecasting method.
This is an open access article under the CC BY-SA license (https://creativecommons.org/licenses/by-sa/4.0/).

Keywords:
Waste Prediction ANFIS Samarinda of the volume of waste increasing daily. Making predictions makes it possible to make earlier preparations and considerations to anticipate if the volume of waste piles increases, which can cause environmental pollution and make residents uncomfortable.
Various analysis of waste handling management and municipal solid waste (MSW) continues to be carried out by researchers with a predictive approach using artificial intelligence. A study in Brisbane, Australia, found that the long short-term memory (LSTM) model outperformed traditional statistical models in forecasting MSW generation, particularly when capturing long-term trends [9]. The LSTM, the ARIMA model, and traditional artificial neural networks (ANN) accuracy reach 0.92, 0.10, and 0.74. The study also found that incorporating demographic data and economic indicators into the LSTM model improved the accuracy of the forecasts. The use of ANN and support vector machines (SVM) in forecasting MSW quantity can potentially improve the efficiency and effectiveness of waste management systems in Johannesburg, South Africa, as well as other urban areas worldwide [10]. In the ANN models, the ten-neuron structure (ANN10) performed best with a determination coefficient (R2) of 99.9%, while in the SVM models, the linear model performed best with an R2 of 98.6%. From the results obtained from the ANN10 model, the total amount of MSW generated per year in the City of Johannesburg is envisaged to get to 1.95 × 106 tonnes in 2050 with an average annual waste of 1.78 x 106 tonnes.
The study analyzes various scenarios and uses a fuzzy technique for order of preference by similarity to the ideal solution (TOPSIS) to forecast MSW generation and evaluate the effectiveness of different waste management strategies to improve municipal MSW planning and forecasting in the Canary archipelago [11]. A modeling study that uses multiple models to forecast MSW generation in China uses multiple regression analysis. The result predicts that MSW generation in China will continue to increase, with an annual growth rate of approximately 3% [12]. The study uses a system dynamics model to simulate plastic waste generation in India over a 50-year period [13]. The model incorporates population growth, economic development, plastic consumption, and waste management practices. The results suggest that without intervention, the amount of plastic waste generated in India will continue to increase rapidly, leading to environmental hazards such as pollution and public health risks. Another study in India uses four different AI models, including ANN, decision trees (DT), random forests (RF), and support vector regression (SVR), to develop a forecast model for MSW generation [14]. The study evaluates the performance of each model and compares their accuracy in forecasting MSW generation. The study results suggest that the ANN model is the most accurate in forecasting MSW generation, followed by the SVR, RF, and DT models.
Waste can be said to be one of the problems faced by many cities around the world, including Indonesia. In Indonesia, according to law number 18 of 2008 concerning waste management, waste is the residue of daily human activities and natural processes in the solid form [15]. Increasing population can cause activities in urban areas to increase, as well as the waste produced. This makes it difficult for the Department of the Environment (DLH) to manage waste, so a predictive study is needed to be appropriately managed. This is crucial for the proper planning and management of waste disposal systems. One method that has shown promise in waste prediction is the Adaptive Neuro-Fuzzy Inference System (ANFIS). ANFIS is a hybrid artificial intelligence technique that combines the advantages of fuzzy logic and neural networks. ANFIS has been successfully used in various applications, including waste prediction, because this method is adaptive, which means that if there is a change in parameter values, it will be connected to existing neurons in getting accurate prediction results [16]. In this context, ANFIS can be used to develop accurate models for waste prediction by analyzing various factors influencing a waste generation, such as population growth, economic development, and waste management practices.
The use of ANFIS in waste prediction can help waste management authorities make informed decisions regarding the design and operation of waste disposal systems, thereby minimizing the impact of waste on the environment and public health. This paper will review the application of ANFIS in waste prediction and its potential for improving waste management practices to assist DLH Samarinda City, East Kalimantan, in adequately organizing the volume of landfill waste. This article consists of the motivation for writing articles in the first part. Second, it describes the working model of the ANFIS method. Third, analysis of the experimental results. The conclusion of the research is at the end.

A. Data Collection
The data used in this study is data on the volume of waste dumps in Samarinda City. Samarinda City is the capital of East Kalimantan, with an area of 718 km2. The period of the waste collection dataset used was from January 2012 -December 2018, with a total of 840 data. The dataset consists of 10 Samarinda City sub-districts, as seen in Table 1.
Meanwhile, population growth continues to increase every year. According to the Central Statistics Agency for Samarinda City, in 2018, the population reached 858,931 people. From 2012-2019 the volume of waste piles in Samarinda City experienced fluctuating changes along with population and industry growth. There are 5:2 and 6:1 data usage ratio scenarios.

B. Data Normalization
Based on the principle of an intelligent system, the data on landfills' volume is first normalized [17]. Data normalization is changing the scale of data within a specific range so that the data has a more balanced distribution and can be processed more effectively by machine learning algorithms. Data normalization helps fix scale issues in data, where some features/variables can have an extensive range of values compared to others, thereby affecting machine learning models' analysis and prediction results. Data normalization can also help speed up the model training process, reducing the number of iterations required and increasing prediction accuracy. The normalization used in this study is min-max normalization. Meanwhile, the normalization formula as in (1) with a range of values [0-1]. After obtaining the predicted results, the data will be denormalized to return the initial values [18].
where, are the results of normalization, ′the data to be normalized, min( )and max( ) are the minimum and maximum values for all data.

C. Adaptive Neuro-Fuzzy Inference System (ANFIS)
ANFIS is an adaptive network based on the Fuzzy Inference System (FIS), using a hybrid learning algorithm procedure. ANFIS can build an input-output based on human knowledge (IF-THEN fuzzy rules) with the proper membership function. In principle, ANFIS parameters can be separated into the premise and consequence parameters that can be adapted to hybrid training. Hybrid learning is carried out in two steps, a forward step and a backward step [19]. The following is the pseudocode for the steps in the ANFIS method. Define the input data and output data 4 Determine the number of membership functions and their parameters for each input variable Calculate the firing strength of each rule based on the degree of membership of each input variable 7 Normalize the firing strength of each rule 8 Calculate the weighted average of the consequent parameters for each rule 9 Calculate the output of the system by summing the weighted averages of the consequent parameters for each rule 10 Use a training algorithm to adjust the antecedent and consequent parameters to minimize the error between the predicted output and the actual output 11 Repeat steps 4-9 for each input-output pair in the training data 12 Test the trained model on new input-output pairs and evaluate its performance.

// End
The network to be implemented uses ANFIS with the Sugeno model. In the Sugeno model, each fuzzy rule predicts the output as a linear function of the input variables. In this experiment, the test parameters used can be seen in Table 2.

D. Predictive Accuracy
MAPE was chosen to measure forecasting accuracy. MAPE is an error measurement that calculates the size of the percentage deviation between actual data and forecast data [20]. The forecasting model will be perfect if it produces a MAPE value that is less than 10% and will be bad if it is above 50%. This study's MAPE score criteria were <10% very good, 10%-20 % good, 20%-50% good enough, and > 50% bad [21]. Meanwhile, the formula for calculating MAPE accuracy measurements can be seen in (2).
where is the amount of data; is the actual data for the period -; is the prediction of the -th period .

III. Results and Discussion
In this part, the experimental findings of the ANFIS approach that has been employed in the process of forecasting the amount of landfill data are described. In order to make an accurate forecast of the findings based on the phases of the experiment, setting the test parameters is crucial. Following that, the testing procedure that was based on data ratios of 5:2 and 6:1 was carried out so that the outcomes of the predictions could be compared, as shown in Table 3. Table 3 demonstrates that virtually all of the MAPE provided by the sub-districts were acceptable because their values were below 50%, with the exception of Samarinda Ulu (K3), which had a value of 56.15%. The MAPE for each individual subdistrict is represented by a separate number in each of the two current ratios. The MAPE in the Sambutan District (K7) is 4.01%, whereas the MAPE in the Sungai Kunjang District (K6) is 31.79%. This represents a ratio of 5:2 between the two districts. The MAPE in North Samarinda (K5) is the lowest, coming in at 3.37%, while the MAPE at K3, which comes in at 56.15%, is the highest. As compared to the 6:1 ratio, the overall findings indicate that the 5:2 ratio has a better average MAPE value (14.61%) than the 6:1 ratio (22.36%). Predictions of waste  Table 4.
According to Table 4, it is known that there is a monthly variation in the total volume of garbage heaps across all of the subdistricts. Both the Sungai Kunjang Subdistrict (K6) and the Loa Janan Ilir Subdistrict (K10) have the same amount of expected monthly volumes over a period of one year, which comes out to 10,269 m3 and 3,047 m3, respectively. The most major garbage mounds, totaling 140,182 m3, are located in the North Samarinda Subdistrict (K5) after one year. In the meanwhile, the Loa Janan Ilir District (K10) has a limited number of garbage heaps that are less than 36,564 m 3 .
From the prediction results using ANFIS for the volume of landfills in 10 sub-districts in Samarinda City in 2019 in Table 3, several suggestions can be submitted to deal with waste problems.
(1) Evaluate the waste management system in each sub-district to reduce the volume of waste heaps produced. This can be done by improving the waste collection, sorting, and processing systems to make them more efficient. (2) Campaign to reduce waste at the source at the individual and community levels. This can be done by educating about good and correct waste management and promoting using recycled products. (3) Improvement of waste management infrastructure in each sub-district, such as constructing appropriate final disposal sites and constructing more effective and efficient waste processing facilities. Thus, it is expected to reduce the volume of landfill waste generated and reduce the negative impact of waste disposal that is not appropriately managed.

IV. Conclusion
Predictions using ANFIS for the volume of landfill waste in 10 sub-districts in Samarinda City, East Kalimantan Province, generally have MAPE that is not bad. The average MAPE for 2018 at a data ratio of 5:2 is 14.61 %, and a data ratio of 6:1 is 22.36 %. The 5:2 data ratio is then used to predict the volume of waste in 2019. From the results of the predictions, each sub-district experienced an increase in the volume of landfills in 2019 because the population and industry have increased. This shows that each sub-district produces predictive values for the volume of waste piles that are not much different. Some suggestions that can be submitted to deal with waste problems in each sub-district are evaluating the waste management system, carrying out waste reduction campaigns, and improving waste management infrastructure. The study's results stated that the ANFIS method was good enough to be used as a predictive method. Future research compares and optimizes the ANFIS method to produce various prediction accuracies.