An Overview of Crop Yield Prediction using Machine Learning Approach

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 10 Issue: 02 | Feb 2023 www.irjet.net p-ISSN: 2395-0072 An Overview of Crop Yield Prediction using Machine Learning Approach Prof. Pritesh A. Patil1, Mr. Pranav Athavale2, Mr. Manas Bothara3, Ms. Siddhi Tambolkar4, Mr. Aditya More5 Dept. of Information Technology AISSMS’s Institute of Information Technology Pune-1, Maharashtra, India ----------------------------------------------------------------------***--------------------------------------------------------------------- Abstract - With the emergence of technologies like draw definitive conclusions or predictive to foretell the machine learning and smart computing, the agriculture future [1]. Subfield of AI, Machine Learning provides a field has seen extensive research in recent years. It is way to computers to capture knowledge from getting harder for farmers to use the land effectively to the dataset—like chess or generating suggestions on earn the most profit in the unique environment because of social networks—without needing to be explicitly the dynamic economics of Agri-produce. Predicting crop taught. Agri-tech and precision farming, which are often yield is a complex task since it depends largely on climatic collectively referred to as "digital agriculture," are variables including soil composition, humidity, and rainfall developing as new fields of research that employ data- as well as area under cultivation and other necessary intensive techniques to boost agricultural output while metrics. Due to a lack of mapping between environmental lowering its environmental effect. In agro - based variables and accompanying algorithms, many present operational contexts, ML has emerged to open up new systems that continuously monitor the aforementioned potential for deciphering, monitoring, and environmental elements provide inaccurate forecasts. It comprehending data-specific processes., along with big creates a negative impact on the accuracy of yield data technology and high-end computers. Data analytics prediction. Machine learning techniques are being sets the groundwork for the development of a diverse deployed to precisely forecast the crop output under array of crop management systems. When data records certain conditions in order to overcome this issue. This are involved, perhaps at the scale of large data, fewer ML project does that by examining and choosing the most operations take place. This is primarily due to the accurate machine learning model. In a circumstance like additional effort required for the data processing this, when there are several crop alternatives, it is crucial activity, which isn't the case with ML models themselves for farmers to prepare their agricultural strategy in [2]. Agriculture sector has a major contribution of advance. The farmer can cultivate in accordance with the 15.87% in India’s GDP in year 2018-19. Also, it is the crop production estimate if he has it in advance. Crop principal source of employment in India employing Yield Prediction (CYP) uses machine learning to assist in nearly 50% of the population [3]. In addition to being a making decisions about which crops to grow and how significant part of the expanding economy, it is crucial much yield they will produce. for our survival. The primary elements affecting agricultural productivity are climate, infestations, and Key Words: Agriculture, Crop yield, Crop prediction, also capacity of harvesting operations. Having reliable Random Forest Regressor, Machine learning crop history data is crucial for controlling agricultural risk [4]. Immoral and illegal methods are being used to 1. INTRODUCTION produce higher yields of less-nutritious hybrid cultivars as the population grows. These methods frequently Machine learning is a rapidly evolving science. Learning degrade soil quality. It damages the ecosystem. As is required when we cannot quickly develop a software weather is getting increasingly fluctuant, farmers are programme to address a particular problem but instead unsure about the crop type, correct sowing times and a need example data or experience. Learning is crucial proper crop strategy. Due to seasonal climatic changes when there is no human knowledge or when individuals and shifts in the availability of essential resources like are unable to articulate their knowledge. Computers are soil, water, and air, the usage of various fertilisers is also configured with software to enhance performance unclear. In this circumstance, the crop yield rate is standards depending on real or fabricated data. Learning steadily declining [5]. is the application of a computer programme to optimise the model's parameters using training data or past Knowing expected yields in advance would help the knowledge. We have a model that has been constructed producer develop a crop strategy. In order to deliver the up to a particular point. The model may be descriptive to most practical of its applications, machine learning is a © 2023, IRJET | Impact Factor value: 8.226 | ISO 9001:2008 Certified Journal | Page 457  International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 10 Issue: 02 | Feb 2023 www.irjet.net p-ISSN: 2395-0072 fast-developing approach that aids decision-making agricultural output of Maharashtra. The most accurate across all sectors. Most modern technologies gain from classifier model used in this study is Random Forest, having their prototypes evaluated before being put to which is followed in accuracy by Logistic Regression and use. The major goal is to increase production in the Naive Bayes. Data such as temperature, humidity, agricultural industry by using ML models. The major rainfall, etc., is fetched using API. The server module emphasis would be on precision agriculture, which puts receives the data that was retrieved from the API. The quality above unfavourable environmental factors [6]. server's database is where the data is kept. The user may Different performance metrics are evaluated in order to give information such as location, area, etc. through the create accurate forecasts and stand by inconsistent mobile application. By completing a single registration, trends in rainfall and temperature. the user may create an account on the mobile app, and all of the submitted information is transferred to the 2. LITERATURE REVIEW server. The RF model on server-end maps the input to the original data and predicts the output. A previous study [7] incorporates a data which contains nutrients and other environmental factors, to forecast Another study by PANDE, SHILPA & RAMESH, PREM & crops. Various feature selection methods and ML models ANMOL, ANMOL & AISHWARYA, B. & ROHILLA, KARUNA are used for CYP. The following variables were examined & SHAURYA, KUMAR [9] says Farming and allure allied in this study: F1 Score, Mean Absolute Error (MAE), subdivisions are certainly the best providers of Logarithmic Loss (LL), Accuracy (ACC), Specificity (S), livelihoods in country India. The importance of Recall (R), Precision (P), and Recall (R) were used to agriculture on a country's GDP cannot be overstated. The evaluate the performance of feature selection and enormous extent of the landmass of the nation fortifies classification algorithms (AUC). Six variables—the it. However, in contrast to universal standards, the average soil and air temperatures, min and max air agricultural output is unsatisfactory. This is one of the temperatures, precipitation, and humidity—are chosen most plausible reasons India's rural farmers have a using the modified elimination of recursive features higher suicide rate. GPS aids in identifying the target (MRFE). Different data splitting validation methods like area. The input from the customer includes the area and (25-75), (30-70), (35-65), (40-60), (45-55), (50-50), (55- soil composition. ML algorithms compile a list of the 45), (60-40), (65-35), (70-30), (75-25) are incorporated most favourable crops, or they forecast the crop yield for and compared against above mentioned accuracy a crop that the consumer has chosen. SVM, MLR, ANN, metrics. Also, variations have been used for feature RF, and KNN are some of the chosen ML algorithms that selection technique in form of MRFE, RFE and Boruta. are used to calculate agricultural production. When Results reveal that, among all the aforementioned k- employed as the ruling class, the Random Forest showed nearest neighbours and bagging classifiers, the Random the best results with 95% accuracy. This design makes Forests Classifier provides the greatest accuracy. The recommendations on when fertiliser should be applied values of the measurements fell as the characteristic to improve production. With regard to data sets from ranges widened. specified area, the suggested model forecasts crop yield. The most important factors in predicting present Different research [8] forecasts agricultural productivity performance are historical data. Several trustworthy using a variety of machine learning techniques. The sources are used to compile historical data. The data sets forecasts made by ML models will help farmers choose are gathered for the regions of Maharashtra and crop which will produce the highest productions. A Karnataka. The information includes a number of technique called data pre-processing is performed to different parameters, including state, district, year, collect the cleaned data from original data collection. As season, crop kind, area under cultivation, productivity, the data is collected in raw form, analysis is not possible. etc. Other databases with state and district details Data is converted into a comprehensible format by using include the soil type as an attribute. The retrieved soil several strategies, such as substituting missing values type column is combined with the primary data set. and null values. The division of training and testing data Similar to how temperature and average rainfall are is the last stage in the data preparation process. Due to added to the primary data sets for the particular the fact that training the model often requires as many location, they are acquired from distinct datasets. The datapoints as feasible, the data typically tend to be data sets have been prepared and cleansed. The mean distributed unevenly. The training dataset, which in this values are used to replace the null values. Before the case makes up 80% of the whole collection, is used to algorithms are run, the categorical attributes are teach ML models how to learn and make accurate transformed into labels. Categorical values in the data predictions. sets are dealt with using the one hot encoding method. By accounting for factors such as temperature, rainfall, The Random Forest regression algorithm was the most acreage, and other features, the study focuses on the accurate of the chosen ones. In order to get the most © 2023, IRJET | Impact Factor value: 8.226 | ISO 9001:2008 Certified Journal | Page 458  International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 10 Issue: 02 | Feb 2023 www.irjet.net p-ISSN: 2395-0072 precise and consistent forecasts, Random Forest should have reliability of prediction above 75%, constructs numerous decision trees and then combines suggesting higher predictive performance. them. The suggested service for fertiliser usage advises the farmer on when to apply the fertiliser. Using Open A study by [11] says that the right crop must be chosen Weather API, the model forecasts rain for a specified before being sowed in order to enhance crop output. It location for the following 14 days. It advises against relies on a number of variables, including the kind of soil using fertilisers if the rainfall is greater than 1.25 mm and its makeup, climate, regional topography, crop and is considered "not safe." output, market pricing, etc. The framework of crop selection, which depends on a lot of variables, has a place A study by Namgiri Suresh, N. V. K. Ramesh, Syed for methods like Decision Trees, K-nearest Neighbors, Inthiyaz, P. Poorna Priya, Kurra Nagasowmika, Kota. V. and Artificial Neural Networks. Crops have been chosen N. Harish Kumar, Mashkoor Shaik and B. N. K. Reddy [10] using machine learning based on how catastrophes like says the majority of India's agricultural products have famines might influence them. Artificial neural networks been severely impacted by the effects of global warming. have been used successfully by researchers to select Considering their output throughout the previous 20 crops based on soil and climate. years. Policymakers and farmers will be able to estimate crop yields early in the harvest by using efficient To satisfy the demands of the soil, maintain its fertility marketing and storage measures. Farmers will be able to levels, and subsequently increase crop output, a plant make the appropriate decisions thanks to this nutrient management system based on machine learning technology because it allows them to know in advance techniques has been developed. It has been suggested to approximate yield of their crops prior to cultivation. The use a crop selection technique called CSM that aids in machine learning algorithm can then be spread when crop selection based on parameters such as yield such a method is implemented with an easy-to-use web- projection. The considerable labour that Indian farmers based graphic programme. The farmer is permitted must endure, such as crop selection, irrigation, and access to the results. However, there are a number of harvesting, might be lessened with an accurate weather protocols or methods for using data analytics to predict forecast. Due to the digital divide, farmers have limited crop yields, and with the help of all those algorithms, we access to the Internet and must rely on the few can forecast agricultural productivity. The Random information about weather forecasts that is accessible. Forest Algorithm is used. The primary goal is to map Data mining is frequently used to address difficulties in data on soil and climate characteristics in the dataset agriculture. Large data sets are analysed using data that contains yield details over the past 12 years. These mining to find valuable classifications and patterns. The factors can help with the prediction of the crops by main objective of the data mining process is to take the utilising various classifiers on the given dataset. As a information from a data collection and organise it so that result, several variables are reviewed, and those that it may be used in other ways. Using the data at hand, this strongly support accurate crop prediction are evaluated. research assesses agricultural yield output. To increase crop production, the crop output was predicted using the Agriculture is the main driver of economic growth in a data mining approach. developing nation. As a nation's population grows, so does its reliance on agriculture, which in turn affects the In a study [12], the suggested method tries to anticipate nation's upcoming economic growth. Crop yield rates or predict crop production by learning from the must be increased to address the hunger need of entire historical data of the farming field. Using machine country. Some biological measures (such as crop variety, learning techniques, the system constructs a forecasting hybrid crops, and insecticides concentration) and model by taking into account many variables such as soil chemical ones (such as fertiliser, urea, and potash use) characteristics, rainfall, temperature, yield, and other are utilised to address this issue. In addition to such things. Here, we employ a variety of machine learning methods, a crop sequencing strategy is required to raise methods, including decision trees, polynomial the crop's web yield rate during the growing season. For regression, and random forests. Predicted accuracy is the purpose of illustrating how it aids farmers in used to evaluate performance. increasing production, the Crop Selection Method (CSM) was utilised as an example. Seasonal crops: Crops can be A study by D. A. Reddy, B. Dadore, and A. Watekar [13] planted at any time during a season, whereas week- emphasizes on the fact that India is one of the countries through crops: Crops are also grown all year long. The that produces the most agricultural goods, yet its farm study demonstrated the usefulness of data mining productivity is still quite low. So that farmers can earn methods for forecasting agricultural yields based on more from the same plot of land with less labour, climatic conditions. Website is user-centric, and all productivity needs to be raised. It provides solutions additional grains and regions selected for the analysis such as providing a recommender utilising an ensemble approach with a high proportion of voting methods © 2023, IRJET | Impact Factor value: 8.226 | ISO 9001:2008 Certified Journal | Page 459  International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 10 Issue: 02 | Feb 2023 www.irjet.net p-ISSN: 2395-0072 using random tree, CHAID, K Nearest Neighbor, and Mean and Standard Deviation for the necessary tuple. Naive Bayes as a classifier to accurately and effectively Calculate the likelihood by comparing the original data advise a good crop based on soil data. Factors are taken sets with the data list that has been summarised. The into account are soil types, soil features, and crop yield biggest probability generated is used for prediction data collecting based on these factors are used to advise based on the outcome. By contrasting the derived class the farmer on the best crop to produce. Precision value with the test data set, the accuracy can be agriculture is one such method that can be used at the estimated. right time to maximise yields and productivity. Another study [15] says the nation's economy benefits One such method used in these research projects is from the field of agriculture. However, it lags behind in assembling. among the numerous machine learning utilising current machine learning technology. Therefore, approaches being applied in this area. Ensembling, all of the latest machine learning technologies and other sometimes referred to as committee methods or model new methods should be familiar to our farmers. These combiners, is a data mining technique that combines the techniques help to increase agricultural productivity. strengths of several models to produce predictions and Agriculture employs a variety of machine learning efficiency that are more accurate than any one model techniques to boost agricultural yield rates. These could produce on its own. Random forests are a method methods can aid in resolving agricultural issues. By for classifying algorithmic rules that uses ensemble examining several approaches, we can also determine learning. the yield accuracy. Thus, by comparing the precision of different crops, we can enhance performance. The use of This system makes use of the majority voting procedure, sensor technologies is widespread in agriculture. This which is the most well-known assembly method. Any study assists in maximising crop output rates. aids in number of base learners may be employed in the voting choosing the appropriate crop for selected land and procedure. There must be at least two base learners. The season. learners are picked so that they complement one other and can teach the others. The possibility of a better The primary objective of agricultural planning is to prediction increases with competition. Using the maximise crop output rates while utilising a certain supplied training data set, the model is trained. Each amount of available land resources. Numerous machine model independently predicts the class when a new learning techniques can aid in increasing crop output sample needs to be classed. The class that the majority of rates. When there is a loss due to unfavourable the students predicted would finally be chosen as the conditions, crop selection can be used to minimise the class label for the new sample. losses. And under favourable circumstances, it can be employed to increase crop yield rates. By maximising In another study [14] Rainfall, perception, production, yield rates, nations' economies are boosted. and temperature data sets are taken into account when building a random forest, a collection of decision trees Crop production may be influenced by the region's that takes into account two-thirds of the records in the natural features, such as riverbeds, hilly terrain, or deep datasets. For correct classification, these decision trees places. such as cloud cover, temperature, rainfall, and are applied to the remaining entries. The test data can be humidity. It might be peaty, saline, sandy, or clay soil. In applied to the resulting training sets for accurate crop soil, you can find copper, potassium, phosphate, yield prediction based on the input attributes. The nitrogen, manganese, iron, calcium, ph level, carbon, and effectiveness of this strategy was examined using the RF different harvesting methods. A number of metrics are algorithm and the dataset. The benefit of the random used for different crops to generate different projections. forest method is that, in contrast to decision tree machine learning algorithms, overfitting is less of a 3. CONCLUSION problem with random forests. No trimming of the random forest is required. Machine learning algorithms The diversity of features that are mostly reliant on the using Random Forest can be generated concurrently. availability of data were reviewed in the current study effort, and CYP was calculated using ML methods that Acquired data sets are converted into csv file format in were distinct from the features. The geological location, accordance with the procedure, after which those data size, and crop features were used to choose the features, sets are loaded. Using a split ratio of either 67 or 33 and these decisions were mostly influenced by the percentage points, or 0.67 or 0.33, the loaded data sets availability of the data collection. However, using more are split into training and test data sets. To categorise features did not necessarily result in better outcomes. As the training data and enable the mapping of attribute a result, testing was done to identify the few best- values to suitable values and list placement. The data performing characteristics that were also included in the sets should then be summarised after determining the research. Neural networks, random forests, KNN © 2023, IRJET | Impact Factor value: 8.226 | ISO 9001:2008 Certified Journal | Page 460  International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 10 Issue: 02 | Feb 2023 www.irjet.net p-ISSN: 2395-0072 regression approaches, and various ML techniques were Approach," 2021 5th International Conference on also employed for the best prediction in the majority of Computing Methodologies and Communication the existing models. According to the study, CNN, LSTM, (ICCMC), 2021, pp. 1066-1071, doi: and DNN algorithms were employed the most frequently, 10.1109/ICCMC51019.2021.9418351. but CYP still needed development. The current study demonstrates a number of current models that [10] N. Suresh et al., "Crop Yield Prediction Using effectively estimate crop yields while taking into account Random Forest Algorithm," 2021 7th International variables like temperature and weather. In the end, the Conference on Advanced Computing and experimental investigation demonstrated how ML can be Communication Systems (ICACCS), 2021, pp. 279- combined with the agricultural domain to enhance crop 282, doi: 10.1109/ICACCS51430.2021.9441871. prediction combining the consequences of various factors on agriculture, however, feature selection still [11] E. Manjula and S. Djodiltachoumy, ``A model for needed to be improved. prediction of crop yield,''Int. J. Comput. Intell. Inform., vol. 6, no. 4, pp. 298–305, 2017. REFERENCES [12] Sangeeta, Shruthi G, “Design And Implementation Of [1] Ethem Alpaydın, Introduction to Machine Learning, Crop Yield Prediction Model In Agriculture”,2020 Second Edition [13] D. A. Reddy, B. Dadore, and A. Watekar, ``Crop [2] Liakos, K.G.; Busato, P.; Moshou, D.; Pearson, S.; recommendation system to maximize crop yield in Bochtis, D. Machine Learning in Agriculture: A ramtek region using machine learning,'' Int. J. Sci. Review. Sensors 2018, 18, 2674. Res. Sci. Technol., vol. 6, no. 1, pp. 485–489 Feb. https://doi.org/10.3390/s18082674 2019. [3] Sabitha; AJEBA, 19(1): 18-31, 2020; Article no. [14] Priya, P., Muthaiah, U., Balamurugan, M.”Predicting AJEBA. 62227 A Study on Sectorial Contribution of Yield of the Crop Using Machine Learning GDP in India from 2010 to 2019 Algorithm”,2015 [4] Jain A. “Analysis of growth and instability in the area, [15] Ramesh Medar,Vijay S, Shweta, “Crop Yield production, yield, and price of rice in India”, Journal Prediction using Machine Learning Techniques”, of Social Change and Development, 2018;2:46-66 2019 [5] Wolfert S, Ge L, Verdouw C, Bogaardt MJ, “Big data in smart farming– a review. Agricultural Systems”, 2017 May 1;153:69-80. [6] Johnson LK, Bloom JD, Dunning RD, Gunter CC, Boyette MD, Creamer NG, “Farmer harvest decisions and vegetable loss in primary production. Agricultural Systems”, 2019 Nov 1;176:102672. [7] S. P. Raja, B. Sawicka, Z. Stamenkovic and G. Mariammal, "Crop Prediction Based on Characteristics of the Agricultural Environment Using Various Feature Selection Techniques and Classifiers," in IEEE Access, vol. 10, pp. 23625- 23641, 2022, doi: 10.1109/ACCESS.2022.3154350. [8] Anakha Venugopal, Aparna S, Jinsu Mani, Rima Mathew, Vinu Williams, 2021, Crop Yield Prediction using Machine Learning Algorithms, INTERNATIONAL JOURNAL OF ENGINEERING RESEARCH & TECHNOLOGY (IJERT) NCREIS – 2021 (Volume 09 – Issue 13) [9] S. M. PANDE, P. K. RAMESH, A. ANMOL, B. R. AISHWARYA, K. ROHILLA and K. SHAURYA, "Crop Recommender System Using Machine Learning © 2023, IRJET | Impact Factor value: 8.226 | ISO 9001:2008 Certified Journal | Page 461